Special Issue “Algorithms for Feature Selection (2nd Edition)”

1. Introduction

This Special Issue focuses on advancing research on algorithms, with a particular emphasis on feature selection techniques. A key focus is the broad subject of scheduling, aiming to include outstanding papers covering both theoretical and practical aspects. The topics comprise evolutionary search techniques for feature selection [1,2,3,4,5], ensemble methods for feature selection [6,7,8,9,10], the problem of feature selection in high-dimensional and time-series data [11,12,13,14,15,16], and the application of textual data [17,18,19,20,21,22] and deep feature selection [23]. Our goal was to bring together some of the best papers from different fields that deal with this subject matter.

A large number of the submitted papers came in response to the call, with ten researchers submitting papers to the Special Issue, each of them approaching a different area of feature selection [24,25,26,27,28]. Through a difficult peer review process, carried out by the invited experts who were there to ensure the quality of the work conducted [29], we are pleased to say that six very good research papers have been accepted for publication in a revised form.

Along with other fields such as scheduling and time-series analysis [30,31,32,33,34,35], the accepted papers also reveal the use of feature selection in emerging fields such as bioinformatics [36,37,38,39,40,41,42] where the selection of the relevant features plays a crucial role in the modeling of genomics. For example, genetic feature selection [43,44,45,46,47,48,49,50] has helped predict cancer better by narrowing down the most important markers [51]. Techniques with neural networks, along with feature importance and dimension reduction, have shown good effectiveness in the optimization of the computational costs and results [52]. These mixed methods have mainly been used in tackling large-scale data challenges, and they have already proved to be successful [53].

Additionally, ensemble methods for feature selection [54,55,56] have been extensively examined in this Special Issue. The methods in question are designed to enhance robustness and lessen the chances of overfitting, which is particularly relevant in the context of big data analytics [57]. The concept of affirmative action is brought to the forefront by one study which shows how ensemble techniques, as opposed to traditional methods, can serve to tackle more data sources [9]. This Special Issue also deals with the challenges of over-dimensional data [58,59], and, in particular, the combination of statistical and machine learning methods has been stressed, though through which dimensionality has been reduced.

2. Special Issue

In the first paper accepted (Neirz et al. 2024) [1], the authors propose a uniqueness testing model based on binomial algorithms. Such a method allows researchers to verify independence between categorical variables in these contexts more effectively in comparison to traditional methodologies such as Fisher’s exact test. The model was confirmed through three case studies investigating the effects of high-pressure processing, ultrasound treatment, and polyphenolic compounds on the structure–function relationship of ancient wheat doughs. The findings support the applicability of computationally simple models in innovation-oriented green product development and the creation of hybrid goods with great nutritional value.

The second paper (Fávero et al. 2024) that was accepted presents a series of nine levels of scenario analysis for the analysis of the different users of a housing space in Alarcos, Cuenca. Research into the problem of planning “affordable housing” has been proposed in terms of demands and scopes of possible solutions. The results verify that the parameters of the models used to explain the problems related to the study of the competition among users of housing spaces provide a good background for decision making. It was the compact output of the scenario analysis chapter which allowed users to administer various policies in the housing market, which was recommended by the housing technicians as an effective tool for the job.

In the third paper accepted (Zulfqar et al. 2024) a novel idea is proposed to use COT-based RGB feature extraction for the detection of lip synchronization irregularities. The method was evaluated for detection performance in a comprehensive set of images, including different variations of input images such as face movement, lighting variation, and other environmental parameters. The paper, moreover, proposed a simple way to unify the scales of static and dynamic features. The experimental results show the effectiveness of the new method and encourage its integration with other existing methods of video content analysis for the detection of non-compliance with synchronization.

In the fourth accepted paper (Mahmood et al. 2024), it has been suggested that the effective use of the aforementioned AI-enabling applications can lead to accelerated realization of student engagement in science through proper scientific tools and methodologies. The application of AI technologies, for instance, can promote a culture of continued learning by providing students with opportunities to work on projects beyond the classroom and facilitating real-time inquiry into their interests. In this regard, I believe the selling of AI applications to students and their involvement as co-creators in the areas of study will be the basic pillars of the future of AI in education.

The fifth accepted paper (Chen et al. 2024) presents a customized detection system for fatigued drivers, which incorporates a variety of data from eye movement, finger pressure, and plantar pressure sensors. The system is a self-learning one that acquires the individual patterns of the driver’s behavioral modifications and can then identify whether the user is feeling tired through the differences, and therefore it provides a clear-cut intervention that is well tailored to road safety enhancement.

The sixth accepted paper (Gafar et al. 2024) describes the RBAVO-DE algorithm, which is a new genetic algorithm based on differential evolution and is specifically designed for gene selection in high-dimensional RNA-Seq datasets. It demonstrates its efficacy by showing better classification accuracy and also feature reduction in 22 cancer datasets, along with being essential in the accuracy of cancer research and genetic marker discovery.

The seventh accepted paper (Ferdous et al. 2024) is about the introduction of AFP-MVFL, which is a multi-view feature learning model dedicated to the scientific identification of antifungal peptides. The combination of sequential and physicochemical properties of amino acids allows the model to achieve the highest performance values, thus offering a very promising tool for the computer-based fight against scientific fungal infections.

The eighth accepted paper (Azimov et al. 2024) discusses the application of machine learning techniques such as CNNs, random forests, and SVMs for the author identification of Azerbaijani texts. The research processed the text features as a variety of known things and reported the best methods of such text classification, representing another contribution to the advancement of computational linguistics.

The ninth accepted paper (Samkunta et al. 2024) proposes a novel method for the extraction of features of hand kinematics of humans based on sparse coding, where a notable improvement in the classification precision can be seen compared to the PCA and random dictionary techniques. This work provides actionable solutions to the problem of time-series data analysis in human movement studies.

The tenth accepted paper (Đurasević et al. 2024) deals with the feature selection process of genetic programming (GP), which is used to build dispatching rules on the basis of the evolution of the phenotype represented in the scheduling problems of machines. Focusing on the role of composite terminal nodes, the article records a 7% increase in performance, with a strong reminder of the importance of the design of the features in the application of GP optimization.