Next Article in Journal
Data Collection Framework for Context-Aware Virtual Reality Application Development in Unity: Case of Avatar Embodiment
Next Article in Special Issue
WiFi FTM and UWB Characterization for Localization in Construction Sites
Previous Article in Journal
SPCBIG-EC: A Robust Serial Hybrid Model for Smart Contract Vulnerability Detection
Previous Article in Special Issue
Indoor Localization Using Uncooperative Wi-Fi Access Points
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Systematic Review

Comprehensive Analysis of Applied Machine Learning in Indoor Positioning Based on Wi-Fi: An Extended Systematic Review †

Faculty of Computer Sciences, Multimedia and Telecommunication, Universitat Oberta de Catalunya, 08018 Barcelona, Spain
Algoritmi Research Center (CALG), Universidade do Minho, 4800-058 Guimarães, Portugal
Authors to whom correspondence should be addressed.
This paper is an extended version of our paper published in the 2021 International Conference on Indoor Positioning and Indoor Navigation (IPIN 2021), Lloret de Mar, Spain, 29 November–2 December 2021.
Sensors 2022, 22(12), 4622;
Received: 11 April 2022 / Revised: 12 June 2022 / Accepted: 15 June 2022 / Published: 19 June 2022
(This article belongs to the Special Issue Indoor Wi-Fi Positioning: Techniques and Systems)


Nowadays, there are a multitude of solutions for indoor positioning, as opposed to standards for outdoor positioning such as GPS. Among the different existing studies on indoor positioning, the use of Wi-Fi signals together with Machine Learning algorithms is one of the most important, as it takes advantage of the current deployment of Wi-Fi networks and the increase in the computing power of computers. Thanks to this, the number of articles published in recent years has been increasing. This fact makes a review necessary in order to understand the current state of this field and to classify different parameters that are very useful for future studies. What are the most widely used machine learning techniques? In what situations have they been tested? How accurate are they? Have datasets been properly used? What type of Wi-Fi signals have been used? These and other questions are answered in this analysis, in which 119 papers are analyzed in depth following PRISMA guidelines.

1. Introduction

The use of outdoor positioning solutions using Global Navigation Satellite Systems (GNSS) technology, such as GPS, GALILEO or GLONASS, is commonplace. Their success lies in the fact that only one receiver (e.g., a cell phone) is needed to obtain the position. However, in closed places (buildings, tunnels, etc.) all of these systems fail, and are unable to obtain a position because the signal cannot penetrate the walls.
To obtain positioning in indoor environments, a technology different from GNSS is needed. Nevertheless, there is not currently an equivalent universal solution. However, in recent years, there has been important progress in many of the technologies used for indoor positioning, including Inertial Positioning [1], Bluetooth [2], Ultrasound [3], Visible Light [4], Wi-Fi [5], etc. These technologies can be applied either individually or together, in what is known as sensor fusion [6,7]. In addition to these “classical” technologies for indoor positioning, promising approximations have recently appeared, such as 5G [8] and Wi-Fi mmWave [9]. Among these possibilities, Wi-Fi-based solutions are very popular, mainly because the infrastructure required for their deployment is already available everywhere, and if it is not, it can be implemented easily and cheaply. For this reason, there are a large number of items based on this technology and the number is growing all the time. In the last few years, there has been a significant increase in the application of machine learning models to enhance the accuracy of indoor positioning. This large volume of works requires a compilation, ordering, and classification of the results in order to assist researchers in selecting appropriate machine learning models for positioning purposes.
Thus, this work has two main contributions: (1) a review of papers published between 2016 and 2021 that use machine learning for indoor positioning, reporting information about the algorithms used, type of article (experimental/simulated), number of Access Point (AP) used, number of radio map reference points used, results obtained, type of signal used, and use or non-use of rooms in experiments; and (2) an analysis of how the main dataset, UJIindoorLoc, has been used in those papers along with the main drawbacks detected when using datasets. The selection of papers in this review was performed following the PRISMA guidelines [10].
This article is an extension of a work presented at the 2021 International Conference on Indoor Positioning and Indoor Navigation (IPIN 2021) [11]. Its novel contents include the following:
  • The current work extends the analyzed period to the last five years, analysing a total of 119 published research works, 57 more than in [11];
  • An analysis of solutions based on Artificial Neural Networks (ANN), Suport Vector Machines (SVM), and Random Forest (RF) is included;
  • A comprehensive analysis of the most widely used public datasets (radio maps) and how they have been integrated in experiments performed by the research community;
  • A discussion of the size of the operational areas considered in experiments performed in the reviewed works;
  • Extended context, discussion, and conclusions.
The rest of this review is organized as follows. Section 2 reviews the existing literature related to indoor positioning. Section 3 describes the methodology used in this paper. Section 4 presents the detailed results in the form of a table. The results of the table are analyzed in Section 5. Finally, Section 6 provides conclusions.

2. Related Work

There are many articles based on Wi-Fi and machine learning algorithms. We found several reviews on this issue, although they answer different questions than those addressed in the current work. For example, [12] is a complete analysis of different indoor positioning articles, however, it is focused on collaborative positioning methods. Collaborative technologies rely on information exchange between different users and/or devices to improve overall performance. The main advantage of this method lies in its infrastructure, which there is less of than in other methods, as well as its low maintenance requirements. Positioning is based on the calculation of data from various sources, such as users and devices, and therefore the main drawback is the need for additional computational resources.
In [13], the authors analyze different articles on indoor positioning; however, they do it at an individual level and do not show any classification or comparative table, although it is a good compilation of articles that use radio, light, or inertial technologies for indoor positioning.
Channel State Information (CSI) for positioning is the focus of the paper in [14], a survey that provides many resources on CSI-based indoor localization methods and includes state-of-the-art algorithms and systems. The authors include a comparative table with fourteen articles using this technology, although only few parameters are analyzed.
Regarding Visible Light Communication for indoor positioning, [15] provides a brief and useful review of ten papers that use machine learning algorithms and visible light solutions in their experiments.
In [16], the authors provide a summary and in-depth analysis of all the wireless technologies used in the field of indoor positioning. Thus, the authors consider works based on the received signal (RSSI or CSI) as well as works that use data such as Time Of Flight (ToF), Angle of Arrival (AoA) or Phase of Arrival (PoA). Their paper includes a review of the different methods used to achieve positioning, such as fingerprinting, multilateration, and triangulation. Finally, they classify the most widely used Machine Learning algorithms and the methods used to filter the received signals. However, it is important to note that the paper is a review of the technologies used, and does not analyze the contributions to the state-of-the-art of every paper individually.
In a paper similar to the previous one, Obeidat et al. [17] review systems not based on radio signals. Thus, the paper reviews positioning through any type of wireless signal as well as optical or magnetic solutions. The authors analyze the different algorithms and techniques used to achieve positioning; however, as in the previous article, they do not sort every paper individually.
Closer to the current review is [18] a survey centered on machine learning algorithms. In this paper, the authors present a compilation of articles based on the application of machine learning algorithms applied to different indoor positioning solutions and classify them by the type of algorithm used. Finally, the authors make a comparative table in which readers can decide which type of algorithm to choose depending on their specific positioning needs (low computational cost, precision, etcetera).
An extensive survey of machine learning techniques for indoor localization and navigation systems is provided in [19], including a deep analysis of all existing algorithms used in this field. The paper is focused on both the algorithms themselves and on different techniques to improve results while working with those algorithms (Data Preprocessing, Interpolating Missing Data, Filtering, etc…). The paper includes reviews of public datasets, performance evaluation parameters used, and other surveys.
Finally, Alhomayani et al. [20] narrow the scope and review fingerprint solutions jointly with deep learning algorithms. Their classification is contains a compilation of the most widely used public Wi-Fi radio maps and a short analysis of every one. However, as in the reviews mentioned previously, their review focuses more on the analysis of the different elements involved in indoor positioning than on analyzing the individual items, which is the focus of the current work.
Thus, as can be seen, many previous works have analyzed how positioning can be achieved using Wi-Fi; nevertheless, there are no previous works, to the best of our knowledge, that have analyzed which machine learning techniques are used, how they are tested, and how much every technique is used. In addition, none of the previous works have analyzed how public datasets (or radio maps) have been integrated into third party research works.

3. Methodology

In this work, in order to analyze the use of machine learning in Wi-Fi solutions to obtain position, the methodology that has been followed is that of a systematic review based on the PRISMA [10] guidelines. The three main steps of this methodology are: (1) to raise the research questions to set the objective of the review; (2) to look for the papers in the chosen digital databases that can answer the research questions; and (3) to establish a set of inclusion and exclusion criteria to finally keep only those papers that fit the research objective. These three steps drive the final selection of the articles that are part of this work.
The research questions are:
Which machine learning algorithms provide the best results in Wi-Fi-based indoor positioning?
What kind of Wi-Fi signal parameters provide the best results?
What are the most commonly used metrics in indoor positioning studies?
Are there substantial differences between simulated and experimental studies?
Which public radio signal maps are the most commonly used in simulations?
To perform queries, the Web of Science and Scopus databases have been chosen; these are reliable sources with sufficient content for an exhaustive review. Figure 1 and Figure 2 show the queries we have used to obtain the scientific papers in the two databases.
The inclusion criteria that selected papers must satisfy are:
Written in English
Coming from a conference or journal article
Dealing with Wi-Fi-based positioning
Positioning through Machine Learning algorithms
Published between 2016 to 2021
The exclusion criteria are:
Workshops and book chapters
Positioning that is not 100% Wi-Fi or is based on Sensor Fusion
Positioning that has part of the work outdoors
Positioning based on classic multilateration (TOA, AOA, etc.)
Positioning that uses a KNN-based algorithm or Particle Filter, as this is not considered Machine Learning
After the list of the papers had been obtained, the next step was to remove duplicates from all the results obtained from the two searches performed in Web of Science and Scopus. With the resulting articles, a first analysis of the title and abstract of each of them was carried out in order to rule out those which failed to meet the inclusion criteria or which met the exclusion criteria. Finally, a full reading was made of the included articles in order to verify whether they met the inclusion criteria. Those that were finally included were analyzed in answering the research questions. The diagram of the different results obtained in each step can be seen in Figure 3.
As can be seen, the original number of papers, after removing duplicates, was 2201. After reviewing them, 119 satisfied the inclusion criteria, and thus are the papers analyzed in the current work.

4. Results

This section presents the results obtained after the analysis of the 119 papers included in this review. The features analyzed regarding the research questions are summarized in Table A1, Table A2, Table A3 and Table A4, which are included in the Appendix. It is important to note the following items:
  • Features not explained in the articles appear as N / A .
  • Articles that include different experiments and/or simulations are grouped together.
  • Articles that do not display a clear metric are marked in the column oError (Other Errors).
  • Articles that are based on or use algorithms different from the main one are marked in the column sAlg (Secondary Algorithm)
The results shown in the tables are discussed and analyzed in the following section.

5. Discussion

In this section, we analyze the results from several points of view: the algorithms used, types of signals used, number of APs and reference points used, metrics, type of experimentation, and most commonly used radio maps.

5.1. Methods: Algorithms and Machine Learning Models

Figure 4 shows the distribution of algorithms. From these results it can be seen that the most commonly used algorithms are those based on ANN. Specifically, there are up to 118 works (around 69 % of the total analyzed works) that use this machine learning model or any of its variants (Deep reinforcement learning (DRL), Extreme learning machine (ELM), Convolutional Neural Networks (CNN), Deep Neural Networks (DNN), Back-Propagation Neural Network (BPNN), Capsule Neural Network (CapsNet), Stacked Denoising Autoencoders (SDA), Variational Autoencoder (VAE), Deep Belief Network (DBN), Recurrent Neural Networks (RNN), Multilayer Perceptron (MLP), Neural Network (NN), Single Multiplicative Neuron (SMN), and Deep Q-Networks (DQN)). Neuronal network-based algorithms are specially appropriate for nonlinear functions, and the fluctuating signal type of Wi-Fi fits perfectly into them.
The number of solutions based on ANN has been growing in recent years. In fact, by the year 2021, 20 out of 24 articles used ANN. In addition, the best result in the analyzed papers (mean error of 0.11   m ) were obtained using a Deep Neural Network to process data from Wi-Fi mmWave signals [9]. However, as we will see later, these results alone do not indicate anything, as factors such as the size of the test area, number of APs, etc. affect these results.
Several papers focus on combining different algorithms in order to choose the one that provides the best results in a particular case [21,22,23], while other papers focus on processing data collected from APs [24,25,26,27]; finally, we found two papers [28,29] that relied on applying a double algorithm, one to approximate the location and another to detail it more precisely from the first approximation.
In the following subsections, we look more deeply into the specifications of NN, SVM, and RF.

5.1.1. Neural Networks

Neural networks are made up of layers of interconnected nodes. Their scheme essentially consists of an input layer, one or more hidden layers, and an output layer. During the training phase, the output is compared with the predicted result and the obtained error is calculated. This error is then propagated through the hidden layers and the weights of the nodes are modified in order to obtain better results. This process is repeated to improve accuracy.
In the articles we analyzed, there is no standard optimum configuration. Researchers perform different tests until they obtain a result that satisfies the two desired properties of accuracy and computation time. In [30], the authors use six layers with 512 nodes in every layer to achieve an accuracy of 2.4 m. However, the authors of [31] instead use only four layers, without specifying the number of nodes, while in [32] only two layers of 50 nodes each are used.
Thus, there is no a standard configuration. Nevertheless, it is important to take into account the difficulty of finding an optimal configuration, as it is influenced by different variables, such as the type of scenario where the experiments are performed, its shape, whether or not there are obstacles, the number of APs used, etc.

5.1.2. Support Vector Machines

The class of algorithms called Suport Vector Machines (SVM) is based on projecting the results on a plane divided into two parts and grouping the results in one of the two parts. Thus, we are talking about a classifier algorithm. In the papers we analysed, we found several different versions of the SVM algorithm. Ref. [33] shows an M-LS-SVM algorithm, which is characterized by the use of linear functions instead of the quadratic functions of the original SVM; the authors obtained an accuracy of 2.7 m. However, [34] used the SVM algorithm directly, obtaining an accuracy of 0.7 m in a similar scenario, and [35] used a SVM algorithm with CSI instead of Received Signal Strength Indicator (RSSI) and obtained an accuracy of 1.909 m in a simpler scenario with no rooms or obstacles.

5.1.3. Random Forest

Random Forest (RF) algorithms are based on the construction of a large number of decision trees to create a learning model. Each decision tree decides a class and the most common class ends up being the final prediction of the model. Its use in indoor positioning has been decreasing, and in the year 2021 no articles were detected that used it. In 2020, there were only three articles that used it, and none of these used scenarios with obstacles to perform the experiments.
The best accuracy found with RF is 1.68 , from Maung et al. [36] in 2020, in a space of 112   m 2 . However, in 2018 the authors of [37] claimed an accuracy of 1.20   m in a space 75   m 2 , and the authors of [38] obtained an accuracy of 0.4033   m in a space of 80   m 2 . From these results it seems that RF is an algorithm suitable for small spaces.

5.1.4. Comparison of Models

The use of one algorithm or another is determined by different factors, such as computational resources, the amount of data to process, and the type of infrastructure (rooms, tables, walls…) where a fingerprinting-based system is to be implemented.
If we focus on computational resources, RF requires fewer resources than SVM algorithms. In fact, SVM-based algorithms tend to be almost unusable on large datasets because the training complexity of SVM is highly dependent on the size of dataset used.
At the level of infrastructure complexity, the situation is similar. SVM algorithms work very well for mitigating the NLOS of signals; therefore, they are ideal in small and complex sites. On the other hand, NNs are more configurable, and their usage can be adjusted for better performance based on lower precision. If not much precision is needed and speed is preferable, the number of nodes and neurons involved in the network can be adjusted [39].
In the case of large spaces, RFs have an advantage over SVMs, because these algorithms are appropriate on models that have been clustered, which is helpful in large scenarios. It is important to note that RFs are Decision Trees optimized to work with large amounts of data. ANNs are particularly suitable in situations where there is noise and multipath propagation, as well as where there are a large number of APs [18].
In summary, in small spaces and with little computational capacity SVM is the best option, while in complex situations and with large datasets, RF and NN are more complex to implement; however, they are more adaptable due to their great configuration capacity.

5.2. Types of Wi-Fi Signal Parameters Used

The most commonly used indoor positioning parameter is based on the RSSI; 114 of the reviewed works used RSSI. The second most used Wi-Fi signal is CSI, which was used by 15 papers.
Wi-Fi is available in many indoor spaces nowadays and is an easily accessible parameter from any device, including mobile and wearable devices. In general, the results obtained with RSSI have an accuracy between 1 and 8 m.
Nevertheless, these apparent good results can be due to the design of the experiments; therefore, these accuracies cannot be generalized or expected in different environments. The elements that drive to these accuracies can be, among others: (1) experiments with small spaces without obstacles and with many reference points, thus avoiding the effect of signal loss when passing through walls and the multipath effect (as explained in [40]); or (2) experiments that use training and validation data with little difference in terms of time and space, or using the data used for training for validation, as in [41].
On the other hand, there are 15 studies that use CSI [32,35,42,43,44,45,46,47,48,49,50,51,52,53] from a Wi-Fi signal, generally with better results in terms of accuracy than those obtained with RSSI. CSI is not widely used because the channel state information is not easy to obtain and requires specific network cards and modifications to the original firmware [52] (i.e., it cannot be used in smartphones). Despite this, we observed a large increase in the use of this parameter. Before the year 2020, only four papers used this parameter. However, in the last two years up to ten papers have used it, as can be seen in Figure 5. One reason for this may be that the RSSI parameter is reaching its limits, and new mechanisms are being explored as they are becoming more present in common everyday devices.
Finally, the Signal-to-Noise Ratio (SNR) parameter is beginning to be used, specifically in two papers [9,54], and in particular in combination with Wi-Fi networks that use mmWave instead of the classical networks that broadcast on traditional frequency channels, i.e., 2.4 GHz and 5 GHz . SNR technologies show better positioning accuracy, and led to the best and the third-best results we found in this review.

5.3. Evaluation Metrics

In order to compare works among themselves, a common evaluation metric is needed. Most works report their results in terms of the average positioning error in different evaluation points, which is the positioning error defined as the Euclidean distance between the actual and estimated positions (Mean Error on the table). Among these, most report the Root Mean Squared Error (RMSE) as well. Other metrics used are the Mean Squared Error (MSE) and the Median Error. Another important metric is the percentage, which is used in one way or another in 36 papers [22,24,25,55,56,57,58,59,60,61,62,63,64,65]. Unfortunately, two articles do not show their results clearly; papers [66,67] only show a graph, however, it is difficult to determine the obtained results from the image.
In analyzing those papers that use metrics recommended by ISO/IEC 18305 [68] (the standard methodology to evaluate indoor localization systems), it can be seen that all of the articles (except those that do not show results) comply with this standard, specifically, mean error, accuracy in one zone or floor, root mean square error, and standard deviation. Figure 6 shows how many papers used each metric.

5.4. Experimental and Full Simulated Results

Regarding whether results were obtained experimentally or via simulation, we found seven papers that presented results from full simulations (with artificially generated data), 40 using public datasets, and 114 that presented empirical results (note that there were articles that perform several experiments and/or simulations).
Four of the papers reporting results based on simulations performed real-world experiments as well [29,45,69,70]. While the authors in [69,70] implemented a simple Log-Distance Path Loss (LDPL) model to generate the RSSI values, Ezzati Khatab et al. [45] included a wall attenuation factor to better model the radio propagation with the LDPL model under Non-Line-of-Sight (NLOS) conditions. In contrast to these, Bai et al. [29] used a more sophisticated Ray Tracing model to generate the RSSI values.
In any event, the results reported in simulations tend to be better than the ones reported in the real-world experiments performed in those papers that performed experiments in both scenarios. Those simulations assuming that Line-of-Sight (LOS) conditions are always met, included a low Gaussian noise, or implemented a simple model, represent an optimistic view of real-world evaluation in one way or another, and therefore the results may be much better in terms of positioning error. This is the case, for instance, in Zhang et al. [70], where the errors in the simulation are 30 % 50 % better than those reported in the real-world experiment.
The empirical results reported are usually better than those obtained with public datasets. There are several reasons for this behavior. In general, researchers have much more knowledge about their own testing areas than those external areas included in public datasets, which impact the selection of the algorithm and its hyperparameters. Performing the experiments in their own facilities enables researchers to select an optimal sub-area for evaluation (e.g., the one with better Wi-Fi coverage or higher density of APs), have custom deployment of APs, or even add additional supporting infrastructure in the operational area. Therefore, public datasets are a more challenging testing scenario for algorithms; in addition, they allow for comparing different algorithms, as they are tested in the same area. Thus, it is important to note the increase in the number of papers that used public datasets in recent years, as can be seen in Figure 7.

5.5. Most Widely Used Public Datasets

Of the experiments reported from 2016, 22 % were performed on public datasets; in 2021 this percentage rises to 39 % , most likely due to the COVID-19 pandemic, although up to six papers from these years do not indicate what type of radio map they used. Thanks to these public datasets, researchers were able to provide useful results while continuing to perform experiments during the pandemic. In addition, these public datasets play a key role in research, as they allow researchers to compare different algorithms tested with the same data.
In the list, the most commonly used public signal map is UJIIndoorLoc [71] in all its variants (different buildings, any floor). It appears in 23 papers, and is clearly the most important, especially in the last two years, when the rest of the maps we found appear only once.
Other radio-maps used are IPIN2016 [72], UTSIndoorLoc [73], JUIndoorLoc [74], Rajen Bhatt [75], Cramariuc [76], Alcala Tutorial 2017 (included in UJIIndoor), WIFINE [77], UJI Library [78], and Tampere [79]. In Table 1 we provide a summary of different attributes of the public datasets used, while Figure 8 shows the evolution of the use of different public datasets over the years. Note that before 2017 there were no public datasets that met the necessary conditions to be used in simulations, thus, we must recognize the recent contribution of these datasets to this field of research.
Because UJIindoorLoc is the most widely used dataset, we performed a deeper analysis into how it has been used by researchers.

UJIIndoor Results Analysis

Due to the large number of articles that performed their tests with UJIIndoorLoc, it is useful to provide a comparison of the different algorithms used on this dataset; results without the mean error have been omitted. Table 2 shows the main papers that used the UJIIndoorLoc dataset.
In analysing these papers, we detected different strategies in researchers’ methods of treating the original UJIIndoor data, resulting in a mean positioning error much below the baseline. From this analysis, we can conclude that the best result obtained by correctly using training and testing from UJIIndoorLoc is [84], with a mean error of 5.64   m .
Despite other works reporting lower positioning error results, these results cannot be directly compared to the baseline as their evaluation was restricted to a small area within the full operational area (a building and/or a floor) and/or the evaluation data contained samples from the original training set, as we show in the following paragraphs. In [41], the author separates the multi-floor and single-floor data to treat them independently, then, from the same dataset, separates 80% for training and 20% for testing. This method can lead to better results, as it is very likely that data taken at almost the same time can be in both the training and test sets. On the contrary, in [80], the authors use the validation component (1111 samples) of the dataset as the test set. The training portion of the dataset is split into training (15,950 samples) and validation (3987 samples) subsets based on an 80:20 split. This is a good practice, as the authors do not mix training and test data.
In [82], the authors selected a subset of the original dataset. In this case, the authors focus on data from only one part of the dataset (building 0) and only choose the strongest RSSI signals. Literally, “In particular, the Building 0 from UJIIndoorLoc dataset is chosen to evaluate EdgeLoc and the top-40 APs (out of a total 520 APs) are selected.” Although AP selection can be performed as an optimisation step in this method, restricting evaluation to just one building makes the results not directly comparable with the baseline method or with other methods that used the full operational area.
The authors in [29] do not specify how the data were used. Literally, “In order to better verify the performance of the algorithm, we also conducted experiments on another widely used positioning dataset UJIIndoorLoc.” However, this algorithm requires a set of evaluation paths to asses the proposed algorithm, which is not provided in the original UJIIndoorLoc dataset. The full details about the data points used and how the evaluation paths were generated are lacking, i.e., the information provided does not enable reproducibility/replicability of the results.
In [83], the authors use the UJIIndoorLoc as an additional experiment alongside their main work. However, the dataset was restricted to a small subset of the UJIIndoorLoc dataset. Literally, “The database from two random phone users (phone id: 13 and 14) in two different buildings (building id: 0 and 1) are used.” In this way, the data to be analyzed and trained on are much simpler and similar, resulting in optimistic performance compared to the baseline method and other solutions using the whole UJIIndoorLoc dataset.
In [85], the authors used two datasets, including UJIIndoorLoc, to assess their proposed model. Despite providing details about the other dataset, they only mention that the UJIIndoorLoc dataset had 21,048 Wi-Fi fingerprints. It seems that the training and evaluation sets were merged into a common superset, which was later split by building ID in order to evaluate the model in three scenarios (buildings). Each of the three sets were split with a ratio of 70:30 to train and evaluate the proposed RDF ensembles.
In [86], the authors randomly split the training set into training and validation sets with a ratio of 80:20. Then, the resulting solution based on CNN was tested over the 1111 evaluation samples, as in the original dataset for the models based on single RSS readings. However, for the method they proposed based on multiple consecutive RSS readings, they had to manipulate the original dataset, splitting the original training set into training, validation, and testing sets with 60%, 20% and 20% of data from the original training set, respectively, i.e., the proposed method was not assessed over an independent test set.
Finally, the original division for training and evaluation provided in UJIIndoorLoc was followed in [81,84]. In [21], the authors do not detail how they trained their model with the UJIIndoorLoc dataset, as they only mention that the dataset contains 21,049 fingerprint samples. Although there is no clear indication about which data were used for training and evaluation, the context provided in the paper suggests that the authors used the evaluation data properly.
To sum up, the UJIIndoorLoc dataset includes a set for training and a set to test the accuracy of an IPS based on Wi-Fi fingerprinting. However, several authors mixed the two sets to apply cross-validation in order to create their own training, validation, and/or test sets, which led to data leakage. In these cases, overly optimistic results were obtained in validation and testing, as the subsets were not independent. In addition, we have observed that full details are often not provided when reporting experiments, which does not enable full reproducibility or replicability of research.

5.6. Experimental Scenarios

Regarding the scenarios in which experiments take place, there is a great diversity of areas. The spaces range from universities to parking lots, stores, residential buildings, etc. As can be seen in Figure 9, 31 articles used an area of less than 100 m 2 , 26 between 100 m 2 and 500 m 2 , 13 between 500 m 2 and 1000 m 2 , and 21 higher than 1000 m 2 . It is important to highlight that in scenarios smaller than 100 m 2 , the use of rooms drops to 41%. In the overall studies, this value is 66%. Although it is true that there are experiments in very small spaces, in general this value indicates that precision is usually prioritized over realistic environments. It should be noted that there are articles with more than one experiment and/or simulation, and every area in these papers is counted here.
Another important aspect of Wi-Fi fingerprinting radio maps is whether they are used in spaces without rooms or with rooms. This simple fact can greatly change the results of a study. Specifically, 103 of the radio maps used have rooms in their experimentation space, 53 were in spaces without any rooms (spaces without walls), and there are 5 papers that do not indicate any such parameter. Finally, it should be noted, again, that most of the works are performed with the focus on obtaining the best results, and not on performing experiments in a realistic environment. For example, changes in RSSI signals are not significant in experiments of short duration and in very delimited spaces.

6. Conclusions

In this paper, we have shown an analysis of the use of machine learning for indoor positioning in Wi-Fi based systems. The starting point has been a systematic review, following the PRISMA guidelines, of the current status of the application of deep learning algorithms applied to indoor positioning using Wi-Fi. Information from 119 articles published between 2016 and 2021 has been extracted and analyzed. In total, 161 simulations or experiments were analyzed.
In this study, we observed a tendency to use Neural Networks in solutions based on the use of Wi-Fi networks. However, we did not find any optimal or standard configuration. In addition to Neural Networks, SVM is widely used as well.
We noted the predominant use of RSSI Wi-Fi signals, although the studies that focus on the CSI are very promising, and are the ones that have obtained the best accuracy; furthermore, in the last year there has been an increase in the number of articles focusing on the use of this information. The only drawback is the difficulty of accessing this information on a Wi-Fi signal.
In analyzing the quality of the results, the Mean Error is the most widely used metric, followed by Accuracy (in percentage). In all cases, the articles analyzed in this review provided results followed the ISO/IEC 18305 guidelines.
Regarding their experiments, we found that most of the papers used empirical results. These papers usually show better results, however, this is generally due to better prepared environments. One important element that we found is that most of the papers prioritize improved results instead of working in a real environment. Thus, although test field sizes range from 6.7   m 2 to 10,000   m 2 , many of these experiments are performed in small work spaces with many reference points and/or APs, or in open spaces without walls, which leads to unrealistic results in everyday environments. Likewise, we found studies in which training and validation data are misused by using repeated values for both sets.
On the other hand, forty papers used public datasets, among which the most popular was UJIIndoorLoc. Using public datasets allows for comparisons to be made between algorithms, as they are tested in the same environment. However, a deep analysis on how UJIIndoorLoc has been used revealed that many authors created their own test and validation data from the training dataset, which leads to overfitting and therefore, to better results than would be obtained with the baseline dataset.
The tables included in this review should be useful for those who want to focus their work based on the size of their work area, choice of machine learning algorithms, and desired accuracy, as well as their choice of the currently most commonly used metrics.

Author Contributions

Conceptualization, V.B.-P., J.T.-S. and A.P.-N.; methodology, V.B.-P., J.T.-S. and A.P.-N.; formal analysis, V.B.-P. and J.T.-S.; data curation, V.B.-P. and A.P.-N.; writing—original draft preparation, V.B.-P.; writing—review and editing, J.T.-S. and A.P.-N.; supervision, J.T.-S. and A.P.-N. All authors have read and agreed to the published version of the manuscript.


A.P.-N. wants to acknowledge the support of GeoLibero CYTED network. J.T.-S. gratefully acknowledge funding from European Union’s Horizon 2020 Research and Innovation programme under Marie Skłodowska Curie grant agreement No. 101023072 (ORIENTATE: Low-cost Reliable Indoor Positioning in Smart Factories, [Accessed on 14 June 2022]).

Institutional Review Board Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.


The following abbreviations are used in this manuscript:
ANNArtificial Neural Networks
AoAAngle of Arrival
APAccess Point
BPNNBack Propagation Neural Network
CapsNetCapsule Neural Network
CNNConvolutional Neural Networks
CSIChannel State Information
DBNDeep Belief Network
DNNDeep Neural Networks
DQNDeep Q-Networks
DRLDeep reinforcement learning
ELMExtreme learning machine
GNSSGlobal Navigation Satellite Systems
MLPMultilayer Perceptron
MSEMean Squared Error
NNNeural Network
PoAPhase of Arrival
RFRandom Forest
RMSERoot Mean Squared Error
RNNRecurrent Neural Networks
RSSIReceived Signal Strength Indicator
SDAStacked Denoising Autoencoders
SMNSingle Multiplicative Neuron
SNRSignal-to-Noise Ratio
SVMSuport Vector Machines
ToFTime Of Flight
VAEVariational Autoencoder

Appendix A. Full Features of Reviewed Articles

Table A1. Summary of reviewed articles.
Table A1. Summary of reviewed articles.
[41]2021pMap16857IPIN2016NDRL 0.92   m RSSI
pMap5891452UTSIndoorLocYDRL 1.72   m RSSI
pMap520993UJIIndoorLocYDRL 3.06   m Only Building B1RSSI
[22]2021pMap9680JUIndoorLocYBayesNetDempster–Shafer Accuracy = 80% between 3 and 3.6 mRSSI
pMap520993UJIIndoorLocYBayesNet Accuracy = 98% in 2 mRSSI
[87]2021exp71161052 m 2 YSISAE (NN) 1.93   m std = 1.34 mRSSI
[44]2021exp132 49.9   m 2 NCNN 1.76   m CSI
exp14540 m 2 NCNN 1.16   m CSI
exp166 48.8 m 2 YCNN 2.54   m CSI
exp11532 m 2 NCNN 0.91   m CSI
[45]2021sim151581160 m 2 YASDELM (ELM) Accuracy = 85,90% in 1 mCSI
exp2247384 m 2 YASDELM (ELM) Accuracy = 77% in 1 mCSI
[88]2021pMap520993UJIIndoorLocYDNNIP Accuracy = 89% building and floorRSSI
[80]2021pMap520993UJIIndoorLocYCHISEL (CNN)autoencoder 6.95 m Accuracy = 99.6% building, 83.97% floorRSSI
[46]2021exp140 131.3 m 2 YBPNNadaptive genetic algorithm Accuracy = 90.47% in 4 mCSI
[30]2021pMap520993UJIIndoorLocYNNELILS (NN) 67% to 78% localization accuraciesRSSI
pMap9680JUIndoorLocYNNELILS (NN) 2.2   m to 2.6   m RSSI
[89]2021pMap3093951TampereYCMDRNN (cnn) 8.26   m std = 1.31 mRSSI
[21]2021pMap520993UJIIndoorLocYCDAE i CNN 12.4 m RSSI
pMap152670Alcala Tutorial 2017NCDAE-CNN 1.05 m RSSI
[89]2021pMap520993UJIIndoorLocYCMDRNN (cnn) 8.26   m std = 1.31 mRSSI
[90]2021exp113303600 m 2 YWiFiNet (cnn) Accuracy = 91.89% in 2 mRSSI
[81]2021pMap520993UJIIndoorLocYDeepLocBox (NN) 9.07 m RSSI
[33]2021exp15150200 m 2 YSVMM-LS 2.7 m RSSI
[47]2021exp1 N / A 14 m 2 NNN 0.18 m CSI
exp2 N / A 18 m 2 NNN 0.03 m CSI
exp2 N / A 6.7 m 2 YNN 0.08 m CSI
[48]2021exp1317 148.5 m 2 YBLS(NN) 2.54 m CSI
exp1176126 m 2 NBLS(NN) 1.48 m CSI
[82]2021exp6132460 m 2 YEdgeloc(CapsNet) 99% under 2 mRSSI
pMap520993UJIIndoorLocYEdgeloc(CapsNet) 7.93 m RSSI
[91]2021exp1210600 m 2 YMLR 4.03 m RSSI
[77]2021exp436654WIFINEYRNN 3.05 m RSSI
[92]2021exp191349360 m 2 NDNN 1.08 m RSSI
[49]2021exp1 17 , 486 CTW 2019 challengeNCNN 0.12 m CSI
[93]2021exp N / A 292600 m 2 YCNN 1.86 m Accuracy = 95% in 5.41 mRSSI
exp N / A 2621360 m 2 YCNN 1.86 m Accuracy = 95% in 5.41 mRSSI
[94]2021exp126806000 m 2 NDNN 3.6 m RSSI
exp121706000 m 2 NDNN 3.7 m RSSI
exp12406000 m 2 NDNN 3.8 m RSSI
[95]2021exp454 69.35 m 2 YANN Accuracy = 13.84% < 0.5 m & 23.07% 0.5 < 1 mRSSI
[50]2020exp32145 m 2 YCNN 1.27 m std = 0.68 mCSI
[36]2020exp4264112 m 2 NRF 1.68 m RSSI
[51]2020exp463 75.6 m 2 NCNN 1.61 m CSI
exp4 N / A 44.8 m 2 NCNN 1.11 m CSI
exp4 N / A 16 m 2 NCNN 0.98 m CSI
[96]2020exp410169 m 2 YCNN 0.98 m RSSI
[54]2020exp53455 m 2 NMLPRegression 0.37 m RMSE = 0.84 mSNR
art: Article; mAlg: Main algorithm used; est: Experimental or pMapulated study; sAlg: Other algorithms used in the study; AP: APs used; mError: Mean Error; rPoint: Reference Points used in offline phase; oError: Other metrics reported in the study; fMap: Size of experimental room or radio-map used; sType: Signal type used; fmRoom: Rooms used in exp/pMap.
Table A2. Summary of reviewed articles.
Table A2. Summary of reviewed articles.
[23]2020pMap520993UJIIndoorLocNKNN, LR, SVM, RF RMSE = 1.87 mRSSI
[55]2020exp6112460 m 2 Ycapsnet 0.68 m RSSI
[31]2020exp8133512 m 2 NDeep Fuzzy Forest 1.36 m RMSE = 1.79 mRSSI
[52]2020exp13250 m 2 NCNN 1.77 m CSI
exp12440 m 2 NCNN 1.16 m CSI
exp16649 m 2 NCNN 2.54 m CSI
[97]2020exp65060 m 2 NRFBernoulli distribution RMSE = 2.50 mRSSI
[98]2020exp25240315 m 2 NRFCo-forest 2.44 m RSSI
exp5 N / A NULLNRF 4.44 m RSSI
[24]2020pMap71000Rajen BhattYMLP Accuracy = 94.4%RSSI
[25]2020pMap520993UJIIndoorLocYCNN Accuracy = 88%RSSI
[99]2020exp195300800 m 2 NDNNHMM 1.22 m RMSE = 1.43 mRSSI
[32]2020exp356 87.75 m 2 NDNNLC 0.78 m std = 1.96 mCSI
[28]2020exp42361148 m 2 YBPNNGA-PSO 0.22 m RSSI
[26]2020exp10102 568.4 m 2 YLSTMLF-D 1.48 m RSSI
exp303532750 m 2 YLSTM 1.75 m RSSI
[27]2020pMap N / A N / A CramariucYSEQ2SEQLSTM 5.5 m RSSI
pMap. N / A N / A CramariucYSEQ2SEQ 3.08 m RSSI
[100]2020pMap N / A N / A IPIN2016YCNN, LSTM 4.93 m RSSI
pMap N / A N / A IPIN2016YCNN, LSTM 5.4 m RSSI
pMap520993UJI LibraryYCNN, LSTM 3.2 m RSSI
pMap520993UJI LibraryYCNN, LSTM 4.98 m RSSI
[56]2020exp522293 m 2 YDNN Accuracy = 95.45% in 3.65 × 3.65 mRSSI
[29]2020exp N / A 1575500 m 2 YRNNDL 3.05 m std = 2.818 mRSSI
pMap520993UJIIndoorLocYRNN 4.92 m std = 3.719 mRSSI
sim4001681 m 2 YRNNDL2.42 m –2.92 m RSSI
[101]2020sim545410,000 m 2 NMLP 3.35 m RSSI
[9]2020exp3725 m 2 YDNNRESNET 0.11 m RMSE = 0.08 mSNR
[102]2020pMap N / A 40UJI LibraryNCNNSVR 2.15 m RSSI
[66]2019exp330540 m 2 NDBNcross entropy and the mean squared NULLRSSI
[34]2019exp259125 m 2 YSVM 0.7 m RSSI
[57]2019exp N / A 206NULLYDNNStacked AutoEncoder Accuracy = 85%RSSI
[35]2019exp1100100 m 2 NSVM 1.9 m std = 0.07 mCSI
[103]2019exp N / A N / A NULLNULLCNN RMSE = 0.31 mRSSI
[104]2019exp1 N / A 63 m 2 YSVM 96.4%RSSI
exp1 N / A 63 m 2 YMLP 96.5%RSSI
[53]2019exp N / A N / A NULLNULLSVM RMSE = 0.42 mCSI
[58]2019exp1683305 m 2 YDNN 2 m RSSI
[59]2019pMap520993UJIIndoorLocYCNN Accuracy = 95.92%RSSI
pMap3093951TampereYCNN Accuracy = 94.13%RSSI
[105]2019exp6300300 m 2 NMEA-BP 0.72 m RSSI
[61]2019exp256741664 m 2 YCNN Accuracy = 95.4% in 4 mRSSI
[62]2019exp541801209 m 2 YRDF Accuracy = 89% at room levelRSSI
[64]2019exp25674300 m 2 YCNN 1.46 m Accuracy = 94% std = 2.24 mRSSI
art: Article; mAlg: Main algorithm used; est: Experimental or pMapulated study; sAlg: Other algorithms used in the study; AP: APs used; mError: Mean Error; rPoint: Reference Points used in offline phase; oError: Other metrics reported in the study; fMap: Size of experimental room or radio-map used; sType: Signal type used; fmRoom: Rooms used in exp/pMap.
Table A3. Summary of reviewed articles.
Table A3. Summary of reviewed articles.
[106]2019exp44280 m 2 NRBFLM 1.42 m RMSE = 1.459 mRSSI
[107]2019exp N / A 300302 m 2 YSVM 4.6 m RSSI
[60]2019exp510NULLNRF Accuracy = 97.5% in 2 mRSSI
[108]2019exp8107512 m 2 YK-ELM RMSE = 1.7123 m std = 2.418 mRSSI
[109]2019exp996560 m 2 YQKMMCC average = 0.76mRSSI
[65]2019pMap520993UJIIndoorLocYRNN Accuracy = 87.41%floor std = 0.83 mRSSI
exp7 N / A 4 RoomsYRNN Accuracy = 95.8% std = 0.60 mRSSI
[83]2019pMap520993UJIIndoorLocYRNN 4.2 m std = 3.2 mRSSI
exp6365336 m 2 YRNN 0.75 m std = 0.64 mRSSI
[110]2019exp9261300 m 2 NBPNN 2.7 m Accuracy = 90%RSSI
[111]2019exp866736 m 2 YSDA 3.7 m Accuracy = 84%RSSI
[112]2019exp14250 m 2 NCNN 0.46 m without obstaclesRSSI
exp14250 m 2 NCNN 1.11 m with some obstaclesRSSI
[113]2019exp11520 m 2 YMLP 1.42 m RSSI
exp11520 m 2 YCNN 1.67 m RSSI
exp115 14.4 m 2 NMLP 1.43 m RSSI
exp115 14.4 m 2 NCNN 1.51 m RSSI
[114]2019exp2589125 m 2 YCNN 3.91 m Accuracy = 84%RSSI
[63]2019pMap N / A N / A NULLYBPNNACO Accuracy = 91.4%RSSI
[115]2019pMap520993UJI LibraryYCNN, GRP 3.6 m 90% less 2mRSSI
[42]2019exp125 26.4 m 2 NBPNNPCA-PD 1.42 m std = 1.1511 mCSI
[84]2019exp N / A 201200 m 2 YMLPSDAE 3.05 m 1dayRSSI
exp N / A 572400 m 2 YMLPSDAE 3.39 m 2 daysRSSI
pMap520993UJIIndoorLocYMLPSDAE 5.64 m 10 daysRSSI
[116]2019pMap520993UJIIndoorLocYVAE RMSE = 4.65 mRSSI
[117]2019exp6491600 m 2 YDNN 0.95 m Open DoorsRSSI
exp6491600 m 2 YDNN 1.26 m Closed DoorsRSSI
[118]2019exp42281200 m 2 YANN 1.22 m RSSI
exp N / A N / A N / A YANN 1.90 m RSSI
[119]2019exp7251728 m 2 NRNNLSTM 1.05 m std = 0.8856 mRSSI
[120]2019exp15714000 m 2 YNNGA 3.47 m RSSI
[121]2019exp4501100 m 2 YBGM 2.9 m RSSI
[122]2019exp12248629 m 2 YDNN 2.64 m RSSI
exp5913965 m 2 NDNN 1.21 m RSSI
[123]2018pMap520993UJIIndoorLocYCNN 95.76% floor levelRSSI
[124]2018pMap71000Rajen BhattYRF 98.3% floor levelRSSI
[125]2018exp2021008250 m 2 YDNN 3.95 m std = 2.72 mRSSI
[126]2018exp16202806 m 2 YSMNPCA 1.85 m std = 1.04 mRSSI
[127]2018pMap520993UJIIndoorLocYDQN 78.79% in 1 mRSSI
[37]2018exp5018075 m 2 YRF 1.29 m 90% in 3 mRSSI
[128]2018exp N / A N / A NULLYDNN 83.6% floor with people, 99.6% withoutRSSI
art: Article; mAlg: Main algorithm used; est: Experimental or pMapulated study; sAlg: Other algorithms used in the study; AP: APs used; mError: Mean Error; rPoint: Reference Points used in offline phase; oError: Other metrics reported in the study; fMap: Size of experimental room or radio-map used; sType: Signal type used; fmRoom: Rooms used in exp/pMap.
Table A4. Summary of reviewed articles.
Table A4. Summary of reviewed articles.
[129]2018pMap N / A N / A UJI LibraryYRNN 2.48 m 99.6% floor levelRSSI
pMap N / A N / A UJI LibraryYLSTM 2.6 m 99.5% floor levelRSSI
[85]2018pMap520993UJIIndoorLocYRDF 6.72 m std = 4.82 mRSSI
[130]2018exp7101 404.5 m 2 YFF-DNN RMSE = 0.32 m, 53.123% in 0.5 mRSSI
[43]2018exp42580 m 2 NRF 0.40 m CSI
[131]2018exp4671664 m 2 YSVM 1.34 m RSSI
[86]2018pMap520993UJIIndoorLocYCNN 2.77 m 100% for floor predictionRSSI
[132]2018exp N / A N / A NULLNULLSVRRBF Kernel 95% in 1.81 mRSSI
[133]2018exp401801209 m 2 YRF 95% accuracy 1.5 × 1.5 mRSSI
[134]2018exp840580 m 2 YRVFL 0.43 m RMSE = 0.5830 mRSSI
[69]2018sim436441 m 2 NRVMPLS 0.84 m RSSI
exp625156 m 2 YRVMPLS 41% in 1 m and 91% in 2 mRSSI
[135]2017exp3110 109.25 m 2 NFF-DNN RMSE = 0.6782 mRSSI
[136]2017exp4 N / A NULLNANN RMSE = 1.1045 mRSSI
exp6 N / A NULLNANN RMSE = 1.2288 mRSSI
[137]2017exp16126304 m 2 YSVM 1.43 m RSSI
[138]2017sim6441100 m 2 NLS-SVM 2.56 m RSSI
[139]2017exp38411600 m 2 YELM 1.91 m RSSI
[140]2017exp286730 m 2 NANN 2.2 m RSSI
[141]2017exp185480NULLYSVM 100% shop levelRSSI
[142]2017pMap520993UJIIndoorLocYDNN 92% floor recognitionRSSI
[143]2017exp848 53.35 m 2 NSVR 86.2% in 1.5 m and 90.4% in 2 mRSSI
[144]2017exp N / A N / A NULLYSVM 97.31% flat and 88.38% floorRSSI
[145]2016exp2284 387.75 m 2 YBPNN 0.98 m RSSI
[146]2016sim425400 m 2 NMLP-ANN 0.27 m std = 0.36 mRSSI
[147]2016sim N / A N / A NULLNULLEB-ANN RMSE = 0.4991 mRSSI
[148]2016exp554150 m 2 YSVR 70% in 5 mRSSI
[149]2016exp161881125 m 2 YANN 1.89 m 90% in 2.971 mRSSI
[70]2016sim1216001600 m 2 NSVR RMSE = 1.42 mRSSI
exp131161000 m 2 YSVR RMSE = 1.8 m, 74% in 2 mRSSI
[150]2016exp N / A 112460 m 2 YSVM 1.2 m RSSI
art: Article; mAlg: Main algorithm used; est: Experimental or pMapulated study; sAlg: Other algorithms used in the study; AP: APs used; mError: Mean Error; rPoint: Reference Points used in offline phase; oError: Other metrics reported in the study; fMap: Size of experimental room or radio-map used; sType: Signal type used; fmRoom: Rooms used in exp/pMap.


  1. Jimenez, A.; Seco, F.; Prieto, J.; Guevara, J. Indoor Pedestrian navigation using an INS/EKF framework for yaw drift reduction and a foot-mounted IMU. In Proceedings of the 2010 7th Workshop on Positioning, Navigation and Communication, Dresden, Germany, 11 March 2010; pp. 135–143. [Google Scholar] [CrossRef]
  2. Faragher, R.; Harle, R. An Analysis of the Accuracy of Bluetooth Low Energy for Indoor Positioning Applications; Institute of Navigation: San Diego, CA, USA, 2014; Volume 1, pp. 201–210. [Google Scholar]
  3. Paredes, J.A.; Álvarez, F.J.; Aguilera, T.; Villadangos, J.M. 3D Indoor Positioning of UAVs with Spread Spectrum Ultrasound and Time-of-Flight Cameras. Sensors 2018, 18, 89. [Google Scholar] [CrossRef][Green Version]
  4. Yoshino, M.; Haruyama, S.; Nakagawa, M. High-accuracy positioning system using visible LED lights and image sensor. In Proceedings of the 2008 IEEE Radio and Wireless Symposium, Orlando, FL, USA, 22–24 January 2008; pp. 439–442. [Google Scholar] [CrossRef]
  5. Kaemarungsi, K.; Krishnamurthy, P. Modeling of indoor positioning systems based on location fingerprinting. In Proceedings of the IEEE INFOCOM 2004, Hong Kong, China, 7–11 March 2004; Volume 2, pp. 1012–1022. [Google Scholar] [CrossRef]
  6. Aguilar Herrera, J.C.; Plöger, P.G.; Hinkenjann, A.; Maiero, J.; Flores, M.; Ramos, A. Pedestrian indoor positioning using smartphone multi-sensing, radio beacons, user positions probability map and IndoorOSM floor plan representation. In Proceedings of the 2014 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Busan, Korea, 27–30 October 2014; pp. 636–645. [Google Scholar] [CrossRef]
  7. De-La-Llana-Calvo, A.; Lázaro-Galilea, J.L.; Gardel-Vicente, A.; Rodríguez-Navarro, D.; Rubiano-Muriel, B.; Bravo-Muñoz, I. Analysis of Multiple-Access Discrimination Techniques for the Development of a PSD-Based VLP System. Sensors 2020, 20, 1717. [Google Scholar] [CrossRef][Green Version]
  8. Horsmanheimo, S.; Lembo, S.; Tuomimaki, L.; Huilla, S.; Honkamaa, P.; Laukkanen, M.; Kemppi, P. Indoor Positioning Platform to Support 5G Location Based Services. In Proceedings of the 2019 IEEE International Conference on Communications Workshops (ICC Workshops), Shanghai, China, 20–24 May 2019; pp. 1–6. [Google Scholar] [CrossRef]
  9. Koike-Akino, T.; Wang, P.; Pajovic, M.; Sun, H.; Orlik, P.P.V. Fingerprinting-Based Indoor Localization with Commercial MMWave WiFi: A Deep Learning Approach. IEEE Access 2020, 8, 84879–84892. [Google Scholar] [CrossRef]
  10. Page, M.J.; McKenzie, J.E.; Bossuyt, P.M.; Boutron, I.; Hoffmann, T.C.; Mulrow, C.D.; Shamseer, L.; Tetzlaff, J.M.; Akl, E.A.; Brennan, S.E.; et al. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. PLoS Med. 2021, 18, e1003583. [Google Scholar] [CrossRef]
  11. Bellavista-Parent, V.; Torres-Sospedra, J.; Perez-Navarro, A. New trends in indoor positioning based on WiFi and machine learning: A systematic review. In Proceedings of the 2021 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Lloret de Mar, Spain, 29 November–2 December 2021; pp. 1–8. [Google Scholar] [CrossRef]
  12. Pascacio, P.; Casteleyn, S.; Torres-Sospedra, J.; Lohan, E.S.; Nurmi, J. Collaborative Indoor Positioning Systems: A Systematic Review. Sensors 2021, 21, 1002. [Google Scholar] [CrossRef] [PubMed]
  13. Kunhoth, J.; Karkar, A.; Al-ma’adeed, S.; Al-Ali, A. Indoor positioning and wayfinding systems: A survey. Hum. Centric Comput. Inf. Sci. 2020, 10, 18. [Google Scholar] [CrossRef]
  14. Liu, W.; Cheng, Q.; Deng, Z.; Chen, H.; Fu, X.; Zheng, X.; Zheng, S.; Chen, C.; Wang, S. Survey on CSI-based Indoor Positioning Systems and Recent Advances. In Proceedings of the 2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Pisa, Italy, 30 September–3 October 2019; pp. 1–8. [Google Scholar] [CrossRef]
  15. Wang, X.; Shen, J. Machine Learning and its Applications in Visible Light Communication Based Indoor Positioning. In Proceedings of the 2019 International Conference on High Performance Big Data and Intelligent Systems (HPBD IS), Shenzhen, China, 9–11 May 2019; pp. 274–277. [Google Scholar] [CrossRef]
  16. Kim Geok, T.; Zar Aung, K.; Sandar Aung, M.; Thu Soe, M.; Abdaziz, A.; Pao Liew, C.; Hossain, F.; Tso, C.P.; Yong, W.H. Review of Indoor Positioning: Radio Wave Technology. Appl. Sci. 2020, 11, 279. [Google Scholar] [CrossRef]
  17. Obeidat, H.; Shuaieb, W.; Obeidat, O.; Abd-Alhameed, R. A Review of Indoor Localization Techniques and Wireless Technologies. Wirel. Pers. Commun. 2021, 119, 289–327. [Google Scholar] [CrossRef]
  18. Nessa, A.; Adhikari, B.; Hussain, F.; Fernando, X.N. A Survey of Machine Learning for Indoor Positioning. IEEE Access 2020, 8, 214945–214965. [Google Scholar] [CrossRef]
  19. Roy, P.; Chowdhury, C. A Survey of Machine Learning Techniques for Indoor Localization and Navigation Systems. J. Intell. Robot. Syst. 2021, 101, 63. [Google Scholar] [CrossRef]
  20. Alhomayani, F.; Mahoor, M.H. Deep learning methods for fingerprint-based indoor positioning: A review. J. Locat. Based Serv. 2020, 14, 129–200. [Google Scholar] [CrossRef]
  21. Qin, F.; Zuo, T.; Wang, X. Ccpos: Wifi fingerprint indoor positioning system based on cdae-cnn. Sensors 2021, 21, 1114. [Google Scholar] [CrossRef]
  22. Roy, P.; Chowdhury, C.; Kundu, M.; Ghosh, D.; Bandyopadhyay, S. Novel weighted ensemble classifier for smartphone based indoor localization. Expert Syst. Appl. 2021, 164, 113758. [Google Scholar] [CrossRef]
  23. Li, L.; Guo, X.; Ansari, N. SmartLoc: Smart Wireless Indoor Localization Empowered by Machine Learning. IEEE Trans. Ind. Electron. 2020, 67, 6883–6893. [Google Scholar] [CrossRef]
  24. Nabati, M.; Navidan, H.; Shahbazian, R.; Ghorashi, S.S.A.; Windridge, D. Using Synthetic Data to Enhance the Accuracy of Fingerprint-Based Localization: A Deep Learning Approach. IEEE Sensors Lett. 2020, 4, 6000204. [Google Scholar] [CrossRef]
  25. Qu, T.; Li, M.; Liang, D. Wireless indoor localization using convolutional neural network. J. Phys. Conf. Ser. 2020, 1633, 012125. [Google Scholar] [CrossRef]
  26. Chen, Z.; Zou, H.; Yang, J.F.; Jiang, H.; Xie, L. WiFi Fingerprinting Indoor Localization Using Local Feature-Based Deep LSTM. IEEE Syst. J. 2020, 14, 3001–3010. [Google Scholar] [CrossRef]
  27. Sun, H.; Zhus, X.; Liu, Y.; Liu, W. Wifi based fingerprinting positioning based on seq2seq model. Sensors 2020, 20, 3767. [Google Scholar] [CrossRef] [PubMed]
  28. Lv, Y.; Liu, W.; Wang, Z.; Zhang, Z. WSN Localization Technology Based on Hybrid GA-PSO-BP Algorithm for Indoor Three-Dimensional Space. Wirel. Pers. Commun. 2020, 114, 167–184. [Google Scholar] [CrossRef]
  29. Bai, S.; Yan, M.; Wan, Q.; He, L.; Wang, X.; Li, J. DL-RNN: An Accurate Indoor Localization Method via Double RNNs. IEEE Sens. J. 2020, 20, 286–295. [Google Scholar] [CrossRef]
  30. Roy, P.; Chowdhury, C. Designing an ensemble of classifiers for smartphone-based indoor localization irrespective of device configuration. Multimed. Tools Appl. 2021, 80, 20501–20525. [Google Scholar] [CrossRef]
  31. Zhang, L.; Chen, Z.; Cui, W.; Li, B.; Chen, C.; Cao, Z.; Gao, K. WiFi-Based Indoor Robot Positioning Using Deep Fuzzy Forests. IEEE Internet Things J. 2020, 7, 10773–10781. [Google Scholar] [CrossRef]
  32. Liu, W.; Chen, H.; Deng, Z.; Zheng, X.; Fu, X.; Cheng, Q. LC-DNN: Local Connection Based Deep Neural Network for Indoor Localization with CSI. IEEE Access 2020, 8, 108720–108730. [Google Scholar] [CrossRef]
  33. Christy Jeba Malar, A.; Deva Priya, M.; Femila, F.; Peter, S.S.; Ravi, V. Wi-Fi Fingerprint Localization Based on Multi-output Least Square Support Vector Regression. Lect. Notes Netw. Syst. 2021, 185, 561–572. [Google Scholar] [CrossRef]
  34. Schmidt, E.; Akopian, D. Indoor Positioning System Using WLAN Channel Estimates as Fingerprints for Mobile Devices. In Proceedings of the SPIE, San Francisco, CA, USA, 8–12 February 2015; Creutzburg, R., Akopian, D., Eds.; SPIE: Bellingham, WA, USA,, 2015; Volume 9411. [Google Scholar] [CrossRef][Green Version]
  35. Yin, L.; Jiang, T.; Deng, Z.; Wang, Z. Improved fingerprint localization algorithm based on channel state information. In Proceedings of the 2019 IEEE 1st International Conference on Civil Aviation Safety and Information Technology (ICCASIT), Kunming, China, 17–19 October 2019; pp. 171–175. [Google Scholar] [CrossRef]
  36. Maung Maung, N.A.; Lwi, B.Y.; Thida, S. An Enhanced RSS Fingerprinting-based Wireless Indoor Positioning using Random Forest Classifier. In Proceedings of the 2020 International Conference on Advanced Information Technologies (ICAIT), Yangon, Myanmar, 4–5 November 2020; pp. 59–63. [Google Scholar] [CrossRef]
  37. Liu, J.; Liu, N.; Pan, Z.; You, X. AutLoc: Deep Autoencoder for Indoor Localization with RSS Fingerprinting. In Proceedings of the 2018 10th International Conference on Wireless Communications and Signal Processing (WCSP), Hangzhou, China, 18–20 October 2018. [Google Scholar] [CrossRef]
  38. Wang, Y.J.; Wang, Y.; Zhang, Y. Indoor Positioning Algorithm for WLAN Based on KFCM-LMC-LSSVM. Jiliang Xuebao Acta Metrol. Sin. 2018, 39, 554–558. [Google Scholar] [CrossRef]
  39. Prinz, A. Computational approaches to neuronal network analysis. Philos. Trans. R. Soc. Biol. Sci. 2010, 365, 2397–2405. [Google Scholar] [CrossRef]
  40. Obeidat, H.A.; Asif, R.; Ali, N.T.; Dama, Y.A.; Obeidat, O.A.; Jones, S.M.; Shuaieb, W.S.; Al-Sadoon, M.A.; Hameed, K.W.; Alabdullah, A.A.; et al. An Indoor Path Loss Prediction Model Using Wall Correction Factors for Wireless Local Area Network and 5G Indoor Networks. Radio Sci. 2018, 53, 544–564. [Google Scholar] [CrossRef]
  41. Dou, F.; Lu, J.; Xu, T.; Huang, C.H.; Bi, J. A Bisection Reinforcement Learning Approach to 3-D Indoor Localization. IEEE Internet Things J. 2021, 8, 6519–6535. [Google Scholar] [CrossRef]
  42. Dang, X.; Ren, J.; Hao, Z.; Hei, Y.; Tang, X.; Yan, Y. A novel indoor localization method using passive phase difference fingerprinting based on channel state information. Int. J. Distrib. Sens. Netw. 2019, 15. [Google Scholar] [CrossRef]
  43. Wang, Y.; Xiu, C.; Zhang, X.; Yang, D. WiFi indoor localization with CSI fingerprinting-based random forest. Sensors 2018, 18, 2869. [Google Scholar] [CrossRef][Green Version]
  44. Han, C.; Xun, W.; Sun, L.; Lin, Z.; Guo, J. DSCP: Depthwise Separable Convolution-Based Passive Indoor Localization Using CSI Fingerprint. Wirel. Commun. Mob. Comput. 2021, 8821129. [Google Scholar] [CrossRef]
  45. Khatab, Z.E.; Gazestani, A.H.; Ghorashi, S.A.; Ghavami, M. A fingerprint technique for indoor localization using autoencoder based semi-supervised deep extreme learning machine. Signal Process. 2021, 181, 107915. [Google Scholar] [CrossRef]
  46. Zhou, M.; Long, Y.; Zhang, W.; Pu, Q.; Wang, Y.; Nie, W.; He, W. Adaptive Genetic Algorithm-Aided Neural Network with Channel State Information Tensor Decomposition for Indoor Localization. IEEE Trans. Evol. Comput. 2021, 25, 913–927. [Google Scholar] [CrossRef]
  47. Gonultas, E.; Lei, E.; Langerman, J.; Huang, H.; Studer, C. CSI-Based Multi-Antenna and Multi-Point Indoor Positioning Using Probability Fusion. IEEE Trans. Wirel. Commun. 2021, 21, 2162–2476. [Google Scholar] [CrossRef]
  48. Wu, C.; Qiu, T.; Zhang, C.; Qu, W.; Wu, D.O. Ensemble Strategy Utilizing a Broad Learning System for Indoor Fingerprint Localization. IEEE Internet Things J. 2021, 9, 3011–3022. [Google Scholar] [CrossRef]
  49. Cerar, G.; Svigelj, A.; Mohorcic, M.; Fortuna, C.; Javornik, T. Improving CSI-based Massive MIMO indoor positioning using convolutional neural network. In Proceedings of the 2021 Joint European Conference on Networks and Communications and 6G Summit, EuCNC/6G Summit 2021, Porto, Portugal, 8–11 June 2021; pp. 276–281. [Google Scholar] [CrossRef]
  50. Zhang, Z.; Lee, M.; Choi, S. Deep Learning-based Indoor Positioning System Using Multiple Fingerprints. Int. Conf. ICT Converg. 2020, 2020, 491–493. [Google Scholar] [CrossRef]
  51. Xiao, Y.; Cui, Z.; Lu, X.; Wang, H. A passive Indoor Localization with Convolutional Neural Network Approach. In Proceedings of the 2020 Chinese Automation Congress (CAC), Shanghai, China, 6–8 November 2020; pp. 1140–1145. [Google Scholar] [CrossRef]
  52. Xun, W.; Sun, L.; Han, C.; Lin, Z.; Guo, J. Depthwise Separable Convolution based Passive Indoor Localization using CSI Fingerprint. In Proceedings of the 2020 IEEE Wireless Communications and Networking Conference (WCNC), Seoul, Korea, 25–28 May 2020. [Google Scholar] [CrossRef]
  53. Ma, C.; Yang, M.; Jin, Y.; Wu, K.; Yan, J. A new indoor localization algorithm using received signal strength indicator measurements and statistical feature of the channel state information. In Proceedings of the 2019 International Conference on Computer, Information and Telecommunication Systems (CITS), Beijing, China, 28–31 August 2019. [Google Scholar] [CrossRef]
  54. Vashist, A.; Bhanushali, D.R.; Relyea, R.; Hochgraf, C.; Ganguly, A.; Manoj, P.S.; Ptucha, R.; Kwasinski, A.; Kuhl, M.E. Indoor wireless localization using consumer-grade 60 GHz equipment with machine learning for intelligent material handling. In Proceedings of the 2020 IEEE International Conference on Consumer Electronics (ICCE), Las Vegas, NV, USA, 4–6 January 2020. [Google Scholar] [CrossRef]
  55. Ye, Q.; Fan, X.; Fang, G.; Bie, H.; Song, X.; Shankaran, R. CapsLoc: A Robust Indoor Localization System with WiFi Fingerprinting Using Capsule Networks. In Proceedings of the ICC 2020-2020 IEEE International Conference on Communications (ICC), Dublin, Ireland, 7–11 June 2020. [Google Scholar] [CrossRef]
  56. Giney, S.; Erdogan, A.; Aktas, M.; Ergun, M. Wi-Fi Based Indoor Positioning System with Using Deep Neural Network. In Proceedings of the 2020 43rd International Conference on Telecommunications and Signal Processing (TSP), Milan, Italy, 7–9 July 2020; pp. 225–228. [Google Scholar] [CrossRef]
  57. Mannoubi, S.; Touati, H. Deep Neural Networks for Indoor Localization Using WiFi Fingerprints. Lect. Notes Comput. Sci. 2019, 11557, 247–258. [Google Scholar] [CrossRef]
  58. Malik, R.F.; Gustifa, R.; Farissi, A.; Stiawan, D.; Ubaya, H.; Ahmad, M.R.; Khirbeet, A.S. The Indoor Positioning System Using Fingerprint Method Based Deep Neural Network. In IOP Conference Series: Earth and Environmental Science; IOP Publishing: Bristol, UK, 2019; Volume 248. [Google Scholar] [CrossRef]
  59. Song, X.; Fan, X.; He, X.; Xiang, C.; Ye, Q.; Huang, X.; Fang, G.; Chen, L.L.; Qin, J.; Wang, Z. Cnnloc: Deep-learning based indoor localization with wifi fingerprinting. In Proceedings of the 2019 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI), Leicester, UK, 19–23 August 2019; pp. 589–595. [Google Scholar] [CrossRef]
  60. Lee, S.; Kim, J.; Moon, N. Random forest and WiFi fingerprint-based indoor location recognition system using smart watch. Hum. Centric Comput. Inf. Sci. 2019, 9, 6. [Google Scholar] [CrossRef]
  61. Haider, A.; Wei, Y.; Liu, S.; Hwang, S.H. Pre- and post-processing algorithms with deep learning classifier for Wi-Fi fingerprint-based indoor positioning. Electronics 2019, 8, 195. [Google Scholar] [CrossRef][Green Version]
  62. Akram, B.A.; Akbar, A.H. Wi-Fi Fingerprinting Based Room Level Indoor Localization Framework Using Ensemble Classifiers. Mehran Univ. Res. J. Eng. Technol. 2019, 38, 151–174. [Google Scholar] [CrossRef]
  63. Chen, J.; Dong, C.; He, G.; Zhang, X. A method for indoor Wi-Fi location based on improved back propagation neural network. Turk. J. Electr. Eng. Comput. Sci. 2019, 27, 2511–2525. [Google Scholar] [CrossRef]
  64. Sinha, R.S.; Hwang, S.H. Comparison of CNN applications for rssi-based fingerprint indoor localization. Electronics 2019, 8, 989. [Google Scholar] [CrossRef][Green Version]
  65. Turabieh, H.; Sheta, A. Cascaded layered recurrent neural network for indoor localization in wireless sensor networks. In Proceedings of the 2019 2nd International Conference on New Trends in Computing Sciences, Amman, Jordan, 9–11 October 2019; pp. 296–301. [Google Scholar] [CrossRef]
  66. Hsu, C.S.; Chen, Y.S.; Juang, T.Y.; Wu, Y.T. An adaptive Wi-Fi indoor localisation scheme using deep learning. In Proceedings of the 2018 IEEE Asia-Pacific Conference on Antennas and Propagation, Auckland, New Zealand, 5–8 August 2018. [Google Scholar] [CrossRef]
  67. Lian, L.; Xia, S.; Zhang, S.; Wu, Q.; Jing, C. Improved Indoor positioning algorithm using KPCA and ELM. In Proceedings of the 2019 11th International Conference on Wireless Communications and Signal Processing (WCSP), Xi’an, China, 23–25 October 2019. [Google Scholar] [CrossRef]
  68. ISO/IEC 18305:2016; Information Technology—Real Time Locating Systems—Test and Evaluation of Localization and Tracking Systems. ISO: Geneva, Switzerland, 2022. Available online: (accessed on 10 April 2022).
  69. Chen, C.; Wang, Y.; Zhang, Y.; Zhai, Y. Indoor positioning algorithm based on nonlinear PLS integrated with RVM. IEEE Sens. J. 2018, 18, 660–668. [Google Scholar] [CrossRef]
  70. Zhang, Y.; Dong, L.; Lai, L.; Hu, L. Study of Indoor Positioning Method Based on Combination of Support Vector Regression and Kalman Filtering. Int. J. Future Gener. Commun. Netw. 2016, 9, 201–214. [Google Scholar] [CrossRef]
  71. Torres-Sospedra, J.; Montoliu, R.; Martínez-Usó, A.; Avariento, J.P.; Arnau, T.J.; Benedito-Bordonau, M.; Huerta, J. UJIIndoorLoc: A new multi-building and multi-floor database for WLAN fingerprint-based indoor localization problems. In Proceedings of the 2014 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Busan, Korea, 27–30 October 2014; pp. 261–270. [Google Scholar] [CrossRef]
  72. Montoliu, R.; Sansano, E.; Torres-Sospedra, J.; Belmonte, O. IndoorLoc platform: A public repository for comparing and evaluating indoor positioning systems. In Proceedings of the 2017 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Sapporo, Japan, 18–21 September 2017; pp. 1–8. [Google Scholar] [CrossRef]
  73. Song, X.; Fan, X.; Xiang, C.; Ye, Q.; Liu, L.; Wang, Z.; He, X.; Yang, N.; Fang, G. A Novel Convolutional Neural Network Based Indoor Localization Framework With WiFi Fingerprinting. IEEE Access 2019, 7, 110698–110709. [Google Scholar] [CrossRef]
  74. Roy, P.; Chowdhury, C.; Ghosh, D.; Bandyopadhyay, S. JUIndoorLoc: A Ubiquitous Framework for Smartphone-Based Indoor Localization Subject to Context and Device Heterogeneity. Wirel. Pers. Commun. 2019, 106, 739–762. [Google Scholar] [CrossRef]
  75. Rohra, J.G.; Perumal, B.; Narayanan, S.J.; Thakur, P.; Bhatt, R.B. User Localization in an Indoor Environment Using Fuzzy Hybrid of Particle Swarm Optimization & Gravitational Search Algorithm with Neural Networks. In Proceedings of the Sixth International Conference on Soft Computing for Problem Solving, Patiala, India, 23–24 December 2016; pp. 286–295. [Google Scholar] [CrossRef]
  76. Cramariuc, A.; Huttunen, H.; Lohan, E.S. Clustering benefits in mobile-centric WiFi positioning in multi-floor buildings. In Proceedings of the 2016 International Conference on Localization and GNSS (ICL-GNSS), Barcelona, Spain, 28–30 June 2016; pp. 1–6. [Google Scholar] [CrossRef]
  77. Khassanov, Y.; Nurpeiissov, M.; Sarkytbayev, A.; Kuzdeuov, A.; Varol, H.A. Finer-level Sequential WiFi-based Indoor Localization. In Proceedings of the 2021 IEEE/SICE International Symposium on System Integration, SII 2021, Iwaki, Fukushima, Japan, 11–14 January 2021; pp. 163–169. [Google Scholar] [CrossRef]
  78. Mendoza-Silva, G.M.; Richter, P.; Torres-Sospedra, J.; Lohan, E.S.; Huerta, J. Long-Term WiFi Fingerprinting Dataset for Research on Robust Indoor Positioning. Data 2018, 3, 3. [Google Scholar] [CrossRef][Green Version]
  79. Lohan, E.S.; Torres-Sospedra, J.; Leppäkoski, H.; Richter, P.; Peng, Z.; Huerta, J. Wi-Fi Crowdsourced Fingerprinting Dataset for Indoor Positioning. Data 2017, 2, 32. [Google Scholar] [CrossRef][Green Version]
  80. Wang, L.; Tiku, S.; Pasricha, S. CHISEL: Compression-Aware High-Accuracy Embedded Indoor Localization with Deep Learning. IEEE Embed. Syst. Lett. 2021, 14, 23–26. [Google Scholar] [CrossRef]
  81. Laska, M.; Blankenbach, J. Deeplocbox: Reliable fingerprinting-based indoor area localization. Sensors 2021, 21, 2000. [Google Scholar] [CrossRef]
  82. Ye, Q.; Bie, H.; Li, K.C.; Gong, X.F.L.; He, X.; Fang, G. EdgeLoc: A Robust and Real-time Localization System Towards Heterogeneous IoT Devices. IEEE Internet Things J. 2021, 9, 3865–3876. [Google Scholar] [CrossRef]
  83. Hoang, M.T.; Yuen, B.; Dong, X.; Lu, T.; Westendorp, R.; Reddy, K. Recurrent Neural Networks for Accurate RSSI Indoor Localization. IEEE Internet Things J. 2019, 6, 10639–10651. [Google Scholar] [CrossRef][Green Version]
  84. Wang, R.; Li, Z.; Luo, H.; Zhao, F.; Shao, W.; Wang, Q. A robust Wi-Fi fingerprint positioning algorithm using stacked denoising autoencoder and multi-layer perceptron. Remote Sens. 2019, 11, 1293. [Google Scholar] [CrossRef][Green Version]
  85. Akram, B.A.; Akbar, A.H.; Shafiq, O. HybLoc: Hybrid indoor wi-fi localization using soft clustering-based random decision forest ensembles. IEEE Access 2018, 6, 38251–38272. [Google Scholar] [CrossRef]
  86. Ibrahim, M.; Torki, M.; Elnainay, M. CNN based Indoor Localization using RSS Time-Series. In Proceedings of the 2018 IEEE Symposium on Computers and Communications (ISCC), Natal, Brazil, 25–28 June 2018; Volume 2018, pp. 1044–1049. [Google Scholar] [CrossRef]
  87. Bai, J.; Sun, Y.; Meng, W.; Li, C. Wi-Fi Fingerprint-Based Indoor Mobile User Localization Using Deep Learning. Wirel. Commun. Mob. Comput. 2021, 2021, 6660990. [Google Scholar] [CrossRef]
  88. Zhang, J.; Su, Y. A Deep Neural Network Based on Stacked Auto-encoder and Dataset Stratification in Indoor Location. In International Conference on Computational Science; Springer: Cham, Switzerland, 2021; pp. 33–46. [Google Scholar] [CrossRef]
  89. Qian, W.; Lauri, F.; Gechter, F. Supervised and semi-supervised deep probabilistic models for indoor positioning problems. Neurocomputing 2021, 435, 228–238. [Google Scholar] [CrossRef]
  90. Hernández, N.; Parra, I.; Corrales, H.; Izquierdo, R.; Ballardini, A.L.; Salinas, C.; García, I. WiFiNet: WiFi-based indoor localisation using CNNs. Expert Syst. Appl. 2021, 177, 114906. [Google Scholar] [CrossRef]
  91. Sugasaki, M.; Shimosaka, M. Robustifying Wi-Fi localization by Between-Location data augmentation. IEEE Sens. J. 2021, 22, 5407–5416. [Google Scholar] [CrossRef]
  92. Chen, C.Y.; Lai, A.I.; Wu, R.B. Multi-Detector Deep Neural Network for High Accuracy Wi-Fi Fingerprint Positioning. In Proceedings of the 2021 IEEE Topical Conference on Wireless Sensors and Sensor Networks, WiSNeT 2021, San Diego, CA, USA, 17–20 January 2021; pp. 37–39. [Google Scholar] [CrossRef]
  93. Li, D.; Xu, J.; Yang, Z.; Lu, Y.; Zhang, Q.; Zhang, X. Train once, locate anytime for anyone: Adversarial learning based wireless localization. In Proceedings of the IEEE INFOCOM 2021-IEEE Conference on Computer Communications, Vancouver, BC, Canada, 10–13 May 2021. [Google Scholar] [CrossRef]
  94. Oh, S.H.; Kim, J.G. DNN based WiFi positioning in 3GPP indoor office environment. In Proceedings of the 3rd International Conference on Artificial Intelligence in Information and Communication, ICAIIC 2021, Jeju Island, Korea, 13–16 April 2021; pp. 302–306. [Google Scholar] [CrossRef]
  95. Puckdeevongs, A. Indoor Localization using RSSI and Artificial Neural Network. In Proceedings of the Proceeding of the 2021 9th International Electrical Engineering Congress, iEECON 2021, Pattaya, Thailand, 10–12 March 2021; pp. 479–482. [Google Scholar] [CrossRef]
  96. Abkari, S.S.E.; Jilbab, A.; Mhamdi, J.J.E.; El Abkari, S.; Jilbab, A.; El Mhamdi, J. RSS-based Indoor Positioning Using Convolutional Neural Network. Int. J. Online Biomed. Eng. 2020, 16, 82–93. [Google Scholar] [CrossRef]
  97. Gao, J.; Li, X.; DIng, Y.; Su, Q.; Liu, Z. WiFi-Based Indoor Positioning by Random Forest and Adjusted Cosine Similarity. In Proceedings of the 2020 Chinese Control and Decision Conference (CCDC), Hefei, China, 22–24 August 2020; pp. 1426–1431. [Google Scholar] [CrossRef]
  98. Zhao, M.; Qin, D.; Guo, R.; Xu, G. Research on Crowdsourcing network indoor localization based on Co-Forest and Bayesian Compressed Sensing. Ad Hoc Netw. 2020, 105, 102176. [Google Scholar] [CrossRef]
  99. Wang, Y.; Gao, J.; Li, Z.; Zhao, L. Robust and accurate Wi-Fi fingerprint location recognition method based on deep neural network. Appl. Sci. 2020, 10, 321. [Google Scholar] [CrossRef][Green Version]
  100. Wang, R.; Luo, H.; Wang, Q.; Li, Z.; Zhao, F.; Huang, J. A Spatial-Temporal Positioning Algorithm Using Residual Network and LSTM. IEEE Trans. Instrum. Meas. 2020, 69, 9251–9261. [Google Scholar] [CrossRef]
  101. Sun, Z.; Zhang, Y.; Ren, Q. A Reliable Localization Algorithm Based on Grid Coding and Multi-Layer Perceptron. IEEE Access 2020, 8, 60979–60989. [Google Scholar] [CrossRef]
  102. Chen, H.; Wang, B.; Pei, Y.; Zhang, L. A WiFi Indoor Localization Method Based on Dilated CNN and Support Vector Regression. In Proceedings of the 2020 Chinese Automation Congress (CAC), Shanghai, China, 6–8 November 2020; pp. 165–170. [Google Scholar] [CrossRef]
  103. Naveed, M.; Javed, Y.; Bhatti, G.M.; Asif, S. Smart indoor Positioning Model for Deterministic Environment. In Proceedings of the 2019 Sixth HCT Information Technology Trends (ITT), Ras Al Khaimah, United Arab Emirates, 20–21 November 2019; pp. 288–291. [Google Scholar] [CrossRef]
  104. You, M.; Park, S.; Lee, S.H.; Yang, T. Proxy individual positioning via IEEE 802.11 monitor mode and fine-tuned analytics. In Proceedings of the 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), Honolulu, HI, USA, 22–25 September 2019. [Google Scholar] [CrossRef]
  105. Jin, X.; Xie, X.; An, K.; Wang, Q.; Guo, J. LoRa Indoor Localization Based on Improved Neural Network for Firefighting Robot. Commun. Comput. Inf. Sci. 2019, 1143 CCIS, 355–362. [Google Scholar] [CrossRef]
  106. Meng, H.; Yuan, F.; Yan, T.; Zeng, M. Indoor Positioning of RBF Neural Network Based on Improved Fast Clustering Algorithm Combined with LM Algorithm. IEEE Access 2019, 7, 5932–5945. [Google Scholar] [CrossRef]
  107. Rubiani, H.; Fitri, S.; Taufiq, M.; Mujiarto, M. Indoor localization based Wi-Fi signal strength using support vector machine. J. Phys. Conf. Ser. 2019, 1402, 077055. [Google Scholar] [CrossRef]
  108. Wang, H.; Li, J.; Cui, W.; Lu, X.; Zhang, Z.; Sheng, C.; Liu, Q. Mobile Robot Indoor Positioning System Based on K-ELM. J. Sens. 2019, 2019, 7547648. [Google Scholar] [CrossRef]
  109. Xue, N.; Luo, X.; Wu, J.; Wang, W.; Wang, L. On the improvement of positioning accuracy in WiFi-based wireless network using correntropy-based kernel learning algorithms. Trans. Emerg. Telecommun. Technol. 2019, 30, e3614. [Google Scholar] [CrossRef]
  110. Wang, B.; Zhu, H.; Xu, M.; Wang, Z.; Song, X. Analysis and Improvement for Fingerprinting-Based Localization Algorithm Based on Neural Network. In Proceedings of the 2019 15th International Conference on Computational Intelligence and Security, Macao, China, 13–16 December 2019; pp. 82–86. [Google Scholar] [CrossRef]
  111. Zhang, H.; Liu, K.; Shang, Q.; Feng, L.; Chen, C.; Wu, Z.; Guo, S. Dual-band wi-fi based indoor localization via stacked denosing autoencoder. In Proceedings of the 2019 IEEE Global Communications Conference (GLOBECOM), Waikoloa, HI, USA, 9–13 December 2019. [Google Scholar] [CrossRef]
  112. Zhao, B.; Zhu, D.; Xi, T.; Jia, C.; Jiang, S.; Wang, S. Convolutional neural network and dual-factor enhanced variational Bayes adaptive Kalman filter based indoor localization with Wi-Fi. Comput. Netw. 2019, 162, 106864. [Google Scholar] [CrossRef]
  113. Xiang, C.; Zhang, S.; Xu, S.; Chen, X.; Cao, S.; Alexandropoulos, G.C.; Lau, V.K. Robust Sub-Meter Level Indoor Localization with a Single WiFi Access Point-Regression Versus Classification. IEEE Access 2019, 7, 146309–146321. [Google Scholar] [CrossRef]
  114. Liu, Z.; Dai, B.; Wan, X.; Li, X. Hybrid wireless fingerprint indoor localization method based on a convolutional neural network. Sensors 2019, 19, 4597. [Google Scholar] [CrossRef] [PubMed][Green Version]
  115. Zhang, G.; Wang, P.; Chen, H.; Zhang, L. Wireless indoor localization using convolutional neural network and gaussian process regression. Sensors 2019, 19, 2508. [Google Scholar] [CrossRef] [PubMed][Green Version]
  116. Chidlovskii, B.; Antsfeld, L. Semi-supervised variational autoencoder for WiFi indoor localization. In Proceedings of the 2019 International Conference on Indoor Positioning and Indoor Navigation (IPIN), Pisa, Italy, 30 September–3 October 2019. [Google Scholar] [CrossRef]
  117. Lin, W.Y.; Huang, C.C.; Duc, N.T.; Manh, H.N. Wi-Fi Indoor Localization based on Multi-Task Deep Learning. In Proceedings of the 2018 IEEE 23rd International Conference on Digital Signal Processing (DSP), Shanghai, China, 19–21 November 2018. [Google Scholar] [CrossRef]
  118. Farid, Z.; Khan, I.U.; Scavino, E.; Abd Rahman, M.A. A WLAN Fingerprinting Based Indoor Localization Technique via Artificial Neural Network. Int. J. Comput. Sci. Netw. Secur. 2019, 19, 157–165. [Google Scholar]
  119. Elbes, M.; Almaita, E.; Alrawashdeh, T.; Kanan, T.; Alzurbi, S.; Hawashin, B. An Indoor Localization Approach Based on Deep Learning for Indoor Location-Based Services. In Proceedings of the 2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology, Amman, Jordan, 9–11 April 2019; pp. 437–441. [Google Scholar] [CrossRef]
  120. Izidio, D.M.; Do Ferreira, A.P.; Da Barros, E.N. Towards better generalization in WLAN positioning systems with genetic algorithms and neural networks. In Proceedings of the 2019 Genetic and Evolutionary Computation Conference, Prague, Czech Republic, 13–17 July 2019; pp. 1206–1213. [Google Scholar] [CrossRef][Green Version]
  121. Alhammadi, A.; Alraih, S.; Hashim, F.; Rasid, M.F.A. Robust 3d indoor positioning system based on radio map using Bayesian network. In Proceedings of the 2019 IEEE 5th World Forum on Internet of Things, Limerick, Ireland, 15–18 April 2019; pp. 107–110. [Google Scholar] [CrossRef]
  122. Abbas, M.; Elhamshary, M.; Rizk, H.; Torki, M.; Youssef, M. WiDeep: WiFi-based accurate and robust indoor localization system using deep learning. In Proceedings of the 2019 IEEE International Conference on Pervasive Computing and Communications PerCom 2019, Kyoto, Japan, 11–15 March 2019. [Google Scholar] [CrossRef]
  123. Jang, J.W.; Hong, S.N. Indoor Localization with WiFi Fingerprinting Using Convolutional Neural Network. In Proceedings of the 2018 Tenth International Conference on Ubiquitous and Future Networks (ICUFN), Prague, Czech Republic, 3–6 July 2018; pp. 753–758. [Google Scholar] [CrossRef]
  124. Gomes, R.; Ahsan, M.; Denton, A. Random Forest Classifier in SDN Framework for User-Based Indoor Localization. In Proceedings of the 2018 IEEE International Conference on Electro/Information Technology (EIT), Rochester, MI, USA, 3–5 May 2018; pp. 537–542. [Google Scholar] [CrossRef]
  125. Chen, S.; Zhu, Q.; Li, Z.; Long, Y. Deep Neural Network Based on Feature Fusion for Indoor Wireless Localization. In Proceedings of the 2018 International Conference on Microwave and Millimeter Wave Technology (ICMMT), Chengdu, China, 7–11 May 2018. [Google Scholar] [CrossRef]
  126. Basiouny, Y.; Arafa, M.; Sarhan, A.M. Enhancing Wi-Fi fingerprinting for indoor positioning system using single multiplicative neuron and PCA algorithm. In Proceedings of the 2017 12th International Conference on Computer Engineering and Systems (ICCES), Cairo, Egypt, 19–20 December 2017; pp. 295–305. [Google Scholar] [CrossRef]
  127. Dou, F.; Lu, J.; Wang, Z.; Xiao, X.; Bi, J.; Huang, C.H. Top-down indoor localization with Wi-Fi fingerprints using deep Q-network. In Proceedings of the 2018 IEEE 15th International Conference on Mobile Ad Hoc and Sensor Systems (MASS), Chengdu, China, 9–12 October 2018; pp. 166–174. [Google Scholar] [CrossRef]
  128. De Vita, F.; Bruneo, D. A deep learning approach for indoor user localization in smart environments. In Proceedings of the 2018 IEEE International Conference on Smart Computing, Taormina, Italy, 18–20 June 2018; pp. 89–96. [Google Scholar] [CrossRef]
  129. Hsieh, H.Y.; Prakosa, S.W.; Leu, J.S. Towards the Implementation of Recurrent Neural Network Schemes for WiFi Fingerprint-Based Indoor Positioning. In Proceedings of the 2018 IEEE 88th Vehicular Technology Conference (VTC-Fall), Chicago, IL, USA, 27–30 August 2018. [Google Scholar] [CrossRef]
  130. Adege, A.A.; Yen, L.; Lin, H.P.; Yayeh, Y.; Li, Y.R.; Jeng, S.S.; Berie, G. Applying Deep Neural Network (DNN) for large-scale indoor localization using feed-forward neural network (FFNN) algorithm. In Proceedings of the 4th Ieee International Conference on Applied System Innovation 2018, Chiba, Japan, 13–17 April 2018; pp. 814–817. [Google Scholar] [CrossRef]
  131. Wei, Y.; Hwang, S.H.; Lee, S.M. IoT-Aided Fingerprint Indoor Positioning Using Support Vector Classification. In Proceedings of the 2018 International Conference on Information and Communication Technology Convergence (ICTC), Jeju, Korea, 17–19 October 2018; pp. 973–975. [Google Scholar] [CrossRef]
  132. Yan, J.; Zhao, L.; Tang, J.; Chen, Y.; Chen, R.; Chen, L. Hybrid Kernel Based Machine Learning Using Received Signal Strength Measurements for Indoor Localization. IEEE Trans. Veh. Technol. 2018, 67, 2824–2829. [Google Scholar] [CrossRef]
  133. Akram, B.A.; Akbar, A.H.; Kim, K.H. CEnsLoc: Infrastructure-Less Indoor Localization Methodology Using GMM Clustering-Based Classification Ensembles. Mob. Inf. Syst. 2018, 2018, 3287810. [Google Scholar] [CrossRef]
  134. Cui, W.; Zhang, L.; Li, B.; Guo, J.; Meng, W.; Wang, H.; Xie, L. Received Signal Strength Based Indoor Positioning Using a Random Vector Functional Link Network. IEEE Trans. Ind. Inform. 2018, 14, 1846–1855. [Google Scholar] [CrossRef]
  135. Belay, A.; Yen, L.; Renu, S.; Lin, H.P.; Jeng, S.S. Indoor localization at 5 GHz using dynamic machine learning approach (DMLA). In Proceedings of the 2017 International Conference on Applied System Innovation (ICASI), Sapporo, Japan, 13–17 May 2017; pp. 1763–1766. [Google Scholar] [CrossRef]
  136. Amirisoori, S.; Abd Aziz, S.M.D.N.S.N.; Sa’at, N.I.M.; Mohd Noor, N.Q. Enhancing Wi-Fi based indoor positioning using fingerprinting methods by implementing neural networks algorithm in real environment. J. Eng. Appl. Sci. 2017, 12, 4144–4149. [Google Scholar] [CrossRef]
  137. Zhang, L.; Li, Y.; Gu, Y.; Yang, W. An efficient machine learning approach for indoor localization. China Commun. 2017, 14, 141–150. [Google Scholar] [CrossRef]
  138. Ezzati Khatab, Z.; Moghtadaiee, V.; Ghorashi, S.A. A fingerprint-based technique for indoor localization using fuzzy Least Squares Support Vector Machine. In Proceedings of the 2017 Iranian Conference on Electrical Engineering (ICEE), Tehran, Iran, 2–4 May 2017; pp. 1944–1949. [Google Scholar] [CrossRef]
  139. Zhang, J.; Sun, J.; Wang, H.; Xiao, W.; Tan, L. Large-scale WiFi indoor localization via extreme learning machine. In Proceedings of the 2017 36th Chinese Control Conference (CCC), Dalian, China, 26–28 July 2017; pp. 4115–4120. [Google Scholar] [CrossRef]
  140. Pahlavani, P.; Gholami, A.; Azimi, S. An indoor positioning technique based on a feed-forward artificial neural network using Levenberg-Marquardt learning method. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2017, 42, 435–440. [Google Scholar] [CrossRef][Green Version]
  141. Rezgui, Y.; Pei, L.; Chen, X.; Wen, F.; Han, C. An Efficient Normalized Rank Based SVM for Room Level Indoor WiFi Localization with Diverse Devices. Mob. Inf. Syst. 2017, 2017, 6268797. [Google Scholar] [CrossRef][Green Version]
  142. Nowicki, M.; Wietrzykowski, J. Low-effort place recognition with WiFi fingerprints using deep learning. In International Conference Automation; Springer: Cham, Switzerland, 2017. [Google Scholar] [CrossRef][Green Version]
  143. Zhao, J.; Wang, J. WiFi indoor positioning algorithm based on machine learning. In Proceedings of the 2017 7th IEEE International Conference on Electronics Information and Emergency Communication (ICEIEC), Macau, China, 21–23 July 2017; pp. 279–283. [Google Scholar] [CrossRef]
  144. Chriki, A.; Touati, H.; Snoussi, H. SVM-based indoor localization in Wireless Sensor Networks. In Proceedings of the 2017 13th International Wireless Communications and Mobile Computing Conference (IWCMC), Valencia, Spain, 26–30 June 2017; pp. 1144–1149. [Google Scholar] [CrossRef]
  145. Abdullah, O.; Abdel-Qader, I. A PNN- Jensen-Bregman Divergence symmetrization for a WLAN Indoor Positioning System. IEEE Int. Conf. Electro Inf. Technol. IEEE Comput. Soc. 2016, 2016, 362–367. [Google Scholar] [CrossRef]
  146. Saleem, F.; Wyne, S. Wlan–Based Indoor Localization Using Neural Networks. J. Electr. Eng. 2016, 67, 299–306. [Google Scholar] [CrossRef][Green Version]
  147. Ibrahim, A.; Rahim, S.K.A.; Mohamad, H. Performance evaluation of RSS-based WSN indoor localization scheme using artificial neural network schemes. In Proceedings of the 2015 IEEE 12th Malaysia International Conference on Communications, Kuching, Malaysia, 23–25 November 2015; pp. 300–305. [Google Scholar] [CrossRef]
  148. Li, L.; Xiang, M.; Zhou, M.; Tian, Z.; Tang, Y. PCA based hybrid hyperplane margin clustering and regression for indoor WLAN localization. In Proceedings of the 2015 10th International Conference on Communications and Networking in China Chinacom, Shanghai, China, 15–17 August 2015; pp. 377–381. [Google Scholar] [CrossRef]
  149. Li, N.; Chen, J.; Yuan, Y.; Tian, X.; Han, Y.; Xia, M. A Wi-Fi Indoor Localization Strategy Using Particle Swarm Optimization Based Artificial Neural Networks. Int. J. Distrib. Sens. Netw. 2016, 2016, 12. [Google Scholar] [CrossRef][Green Version]
  150. Wu, Z.; Fu, K.; Jedari, E.; Shuvra, S.R.; Rashidzadeh, R.; Saif, M. A Fast and Resource Efficient Method for Indoor Positioning Using Received Signal Strength. IEEE Trans. Veh. Technol. 2016, 65, 9747–9758. [Google Scholar] [CrossRef]
Figure 1. Query for Scopus.
Figure 1. Query for Scopus.
Sensors 22 04622 g001
Figure 2. Query for Web of Science.
Figure 2. Query for Web of Science.
Sensors 22 04622 g002
Figure 3. PRISMA flow diagram.
Figure 3. PRISMA flow diagram.
Sensors 22 04622 g003
Figure 4. Most widely used algorithms and Machine Learning models.
Figure 4. Most widely used algorithms and Machine Learning models.
Sensors 22 04622 g004
Figure 5. Evolution of the types of signal used.
Figure 5. Evolution of the types of signal used.
Sensors 22 04622 g005
Figure 6. Metrics used.
Figure 6. Metrics used.
Sensors 22 04622 g006
Figure 7. Evolution of experimental vs. simulated studies.
Figure 7. Evolution of experimental vs. simulated studies.
Sensors 22 04622 g007
Figure 8. Evolution of the use of public datasets over the years.
Figure 8. Evolution of the use of public datasets over the years.
Sensors 22 04622 g008
Figure 9. Size of scenarios used in experiments (in square meters).
Figure 9. Size of scenarios used in experiments (in square meters).
Sensors 22 04622 g009
Table 1. Public dataset summary.
Table 1. Public dataset summary.
Public Radio MapYearSizeAPsrPointsOthers
UJIIndoorLoc2014110,000 m 2 520993three buildings with four or five floors depending on the building.
IPIN20162016150 m 2 16857a university corridor
UTSIndoorLoc201944,000 m 2 5891452a building with sixteen floors, including three basement levels
JUIndoorLoc20192646 m 2 1722646faculty rooms, classrooms, seminar rooms, research labs, and corridor
Rajen Bhatt20194 rooms71000conference room, kitchen, or indoor sports room
Cramariuc20162 university building6632651data divided into two different University buildings.
WiFine20209000 m 2 43626,418based on 260 trajectories
UJI Library2020 308.4 m 2 448212data taken across fifteen months at the same positions and directions
Tampere201722,570 m 2 9924648882 rooms on six floors
Table 2. Articles that used the UJIIndoorLoc dataset.
Table 2. Articles that used the UJIIndoorLoc dataset.
[41]2021DRL 3.06   m
[80]2021CHISEL (CNN) 6.95   m
[21]2021CNN 12.4   m
[81]2021DeepLocBox (NN) 9.07   m
[82]2021Edgeloc(CapsNet) 7.93   m
[29]2020RNN 4.91   m
[83]2019RNN 4.2   m
[84]2019MLP 5.64   m
[85]2018RDF 6.72   m
[86]2018CNN—Single RSS vector 10.25   m
CNN—Time Series 2.77   m
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Bellavista-Parent, V.; Torres-Sospedra, J.; Pérez-Navarro, A. Comprehensive Analysis of Applied Machine Learning in Indoor Positioning Based on Wi-Fi: An Extended Systematic Review. Sensors 2022, 22, 4622.

AMA Style

Bellavista-Parent V, Torres-Sospedra J, Pérez-Navarro A. Comprehensive Analysis of Applied Machine Learning in Indoor Positioning Based on Wi-Fi: An Extended Systematic Review. Sensors. 2022; 22(12):4622.

Chicago/Turabian Style

Bellavista-Parent, Vladimir, Joaquín Torres-Sospedra, and Antoni Pérez-Navarro. 2022. "Comprehensive Analysis of Applied Machine Learning in Indoor Positioning Based on Wi-Fi: An Extended Systematic Review" Sensors 22, no. 12: 4622.

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop