# Enhancing Urban Data Analysis: Leveraging Graph-Based Convolutional Neural Networks for a Visual Semantic Decision Support System

^{1}

^{2}

^{3}

^{*}

## Abstract

**:**

## 1. Introduction

#### 1.1. The Crucial Impact of Urban Planning on City Development

#### 1.2. Urban Planning Challenges

#### 1.3. Research Contribution

## 2. Related Work

#### 2.1. Urban Planning Challenges

#### 2.2. Convolutional Neural Networks’ Architecture

#### 2.3. CNNs in Urban Planning

## 3. Premise Outline and Principles

## 4. Problem Data Collection and Preprocessing

#### 4.1. Problem Formulation

#### 4.2. Description of the Dataset and Data Collection

#### 4.3. Graph Preprocessing

## 5. Graph-Based Visual Representation

## 6. Experimental Setup

#### 6.1. Experiment CNN Architectural Overview and Implementation

#### 6.2. CNN Evaluation

- True Positives (TPs): $TP={P}_{c}\cap {P}_{e}$
- True Negatives (TNs): $TN={N}_{c}\cap {N}_{e}^{i}$
- False Positives (FPs): $FP={P}_{c}\cap {N}_{e}^{i}$
- False Negatives (FNs): $FN={N}_{c}\cap {P}_{e}$

- Accuracy:

- Specificity:

- Precision:

- Recall (or Sensitivity):

- F1 measure: The F1 measure is the harmonic mean between Precision and Recall. Its range is [0, 1]. It provides information on how precise the classifier is (how many instances it classifies correctly), as well as how robust it is (if it misses a significant number of instances).

- G-mean: The geometric mean (G-mean) is the root of the product of the class-wise Sensitivity. This measure tries to maximize the Accuracy for each of the classes while keeping the Accuracy values balanced. For binary classification, G-mean is the squared root of the product of the Sensitivity and Specificity.

#### 6.3. AlexNet CNN

#### 6.4. SqueezeNet CNN

- Blue: These blocks represent the input and output layers where the initial image data is fed into the network and the final classification is outputted.
- Orange: These represent ReLU (Rectified Linear Unit Layers.
- Purple: These indicate max-pooling layers used for down-sampling and reducing the spatial dimensions. They help to make the model more robust to variations in the input and reduce overfitting.
- Yellow: These depict connected layers that perform classification based on the features extracted by the previous layers.
- Dark Red: These are softmax used as the last activation function to normalize the output of a network to a probability distribution over predicted output classes.
- Brown: these represent DepthConcatenationLayer: In some network architectures, the outputs of several layers might be concatenated together along the channel dimension.

#### 6.5. VGG-16 CNN

#### 6.6. Comparison with Results from Machine Learning Classifiers

#### 6.7. Discussion

## 7. Conclusions and Future Work

#### 7.1. Conclusions

#### 7.2. Future Work

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Conflicts of Interest

## References

- Ritchie, H.; Roser, M. Urbanization. Our World Data. 2023. Available online: https://ourworldindata.org/urbanization (accessed on 6 February 2024).
- Keith, M.; Birch, E.; Buchoud, N.; Cardama, M.; Cobbett, W.; Cohen, M.; Elmqvist, T.; Espey, J.; Hajer, M.; Hartmann, G.; et al. A new urban narrative for sustainable development. Nat. Sustain.
**2022**, 6, 115–117. [Google Scholar] [CrossRef] - Abd Elrahman, A.S.; Asaad, M. Urban design & urban planning: A critical analysis to the theoretical relationship gap. Ain Shams Eng. J.
**2021**, 12, 1163–1173. [Google Scholar] [CrossRef] - Mahtta, R.; Fragkias, M.; Güneralp, B.; Mahendra, A.; Reba, M.; Wentz, E.A.; Seto, K.C. Urban land expansion: The role of population and economic growth for 300+ cities. NPJ Urban Sustain.
**2022**, 2, 5. [Google Scholar] [CrossRef] - Tékouabou, S.C.K.; Chenal, J.; Azmi, R.; Toulni, H.; Diop, E.B.; Nikiforova, A. Identifying and Classifying Urban Data Sources for Machine Learning-Based Sustainable Urban Planning and Decision Support Systems Development. Data
**2022**, 7, 170. [Google Scholar] [CrossRef] - Cesario, E. Big data analytics and smart cities: Applications, challenges, and opportunities. Front. Big Data
**2023**, 6, 1149402. [Google Scholar] [CrossRef] [PubMed] - Hudson-Smith, A.; Wilson, D.; Gray, S.; Dawkins, O. Urban IoT: Advances, Challenges, and Opportunities for Mass Data Collection, Analysis, and Visualization. In Urban Informatics; Shi, W., Goodchild, M.F., Batty, M., Kwan, M.-P., Zhang, A., Eds.; The Urban Book Series; Springer: Singapore, 2021; pp. 701–719. ISBN 9789811589836. [Google Scholar]
- Silva, J.S.; da Silva, R.M.; Santos, C.A.G. Spatiotemporal impact of land use/land cover changes on urban heat islands: A case study of Paço do Lumiar, Brazil. Build. Environ.
**2018**, 136, 279–292. [Google Scholar] [CrossRef] - Hamilton, W.; Ying, Z.; Leskovec, J. Inductive Representation Learning on Large Graphs. In Advances in Neural Information Processing Systems; Curran Associates, Inc.: Red Hook, NY, USA, 2017; Available online: https://proceedings.neurips.cc/paper_files/paper/2017/hash/5dd9db5e033da9c6fb5ba83c7a7ebea9-Abstract.html (accessed on 1 February 2024).
- Chen, M.; Mao, S.; Liu, Y. Big Data: A Survey. Mob. Netw. Appl.
**2014**, 19, 171–209. [Google Scholar] [CrossRef] - Krizhevsky, A.; Sutskever, I.; Hinton, G.E. ImageNet Classification with Deep Convolutional Neural Networks. In Advances in Neural Information Processing Systems; Curran Associates, Inc.: Red Hook, NY, USA, 2012; Available online: https://proceedings.neurips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html (accessed on 12 November 2023).
- Veličković, P.; Cucurull, G.; Casanova, A.; Romero, A.; Liò, P.; Bengio, Y. Graph Attention Networks. arXiv
**2018**, arXiv:1710.10903. [Google Scholar] - Krichen, M. Convolutional Neural Networks: A Survey. Computers
**2023**, 12, 151. [Google Scholar] [CrossRef] - Huang, B.; Zhao, B.; Song, Y. Urban land-use mapping using a deep convolutional neural network with high spatial resolution multispectral remote sensing imagery. Remote Sens. Environ.
**2018**, 214, 73–86. [Google Scholar] [CrossRef] - Yang, J.; Kwon, Y. Novel CNN-Based Approach for Reading Urban Form Data in 2D Images: An Application for Predicting Restaurant Location in Seoul, Korea. ISPRS Int. J. Geo-Inf.
**2023**, 12, 373. [Google Scholar] [CrossRef] - Chen, Y.; Weng, Q.; Tang, L.; Liu, Q.; Zhang, X.; Bilal, M. Automatic mapping of urban green spaces using a geospatial neural network. GIScience Remote Sens.
**2021**, 58, 624–642. [Google Scholar] [CrossRef] - Bao, H.; Ming, D.; Guo, Y.; Zhang, K.; Zhou, K.; Du, S. DFCNN-Based Semantic Recognition of Urban Functional Zones by Integrating Remote Sensing Data and POI Data. Remote Sens.
**2020**, 12, 1088. [Google Scholar] [CrossRef] - Sideris, N.; Bardis, G.; Voulodimos, A.; Miaoulis, G.; Ghazanfarpour, D. Using Random Forests on Real-World City Data for Urban Planning in a Visual Semantic Decision Support System. Sensors
**2019**, 19, 2266. [Google Scholar] [CrossRef] [PubMed] - Tammina, S. Transfer learning using VGG-16 with Deep Convolutional Neural Network for Classifying Images. Int. J. Sci. Res. Publ.
**2019**, 9, 420. [Google Scholar] [CrossRef] - Alom, M.Z.; Taha, T.M.; Yakopcic, C.; Westberg, S.; Sidike, P.; Nasrin, M.S.; Van Esesn, B.C.; Awwal, A.A.S.; Asari, V.K. The History Began from AlexNet: A Comprehensive Survey on Deep Learning Approaches. arXiv
**2018**, arXiv:1803.01164. [Google Scholar] [CrossRef] - Iandola, F.N.; Han, S.; Moskewicz, M.W.; Ashraf, K.; Dally, W.J.; Keutzer, K. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size. arXiv
**2016**, arXiv:1602.07360. [Google Scholar] [CrossRef] - Research—DataGrandLyon. Available online: https://data.grandlyon.com/portail/en/recherche (accessed on 4 February 2024).
- Kipf, T.N.; Welling, M. Semi-Supervised Classification with Graph Convolutional Networks. arXiv
**2017**, arXiv:1609.02907. [Google Scholar] [CrossRef] - Battiston, F.; Cencetti, G.; Iacopini, I.; Latora, V.; Lucas, M.; Patania, A.; Young, J.-G.; Petri, G. Networks beyond pairwise interactions: Structure and dynamics. Phys. Rep.
**2020**, 874, 1–92. [Google Scholar] [CrossRef] - Buttenfield, B.P.; McMaster, R.B. Map Generalization: Making Rules for Knowledge Representation. Citeseer. 1991. Available online: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=97fdf14eadea9ff72c058d4e4c4b24b6c58b346a (accessed on 12 November 2023).

**Figure 1.**Aspect of the visual representation of the related points of interest: blue circles indicate parking lots, red circles indicate points of touristic interest and green circles indicate bike rental points. The lines connecting them indicate that the Euclidean distance between them is less than 500 m. Street names and locations of Lyon, contained in the map excerpt, appear in French.

**Figure 3.**Representative visualization samples from each category, where the difference in distance, plurality of representative neighboring features, and area are evident. (

**a**) Graph-based visualization of a parking building (positive classification sample); (

**b**) graph-based visualization of a false parking building (negative classification sample). In both samples the central black circle is representative of the site under classification, while the other circles follow the same color scheme adopted in Figure 1. The color of the edges is determined by their weight according to a standard colormap.

**Figure 4.**Training plot for AlexNet. The upper diagram illustrates the accuracy as it is formed after each training epoch, where the blue line represents the training accuracy, and the dark line represents the validation accuracy. The lower diagram depicts the loss, where the orange line represents the training loss and the dark dotted line represents the validation loss.

**Figure 5.**Confusion matrix for AlexNet. The colors in all matrices represent the values of the predictions. Darker colors (blue) indicate higher percentages, while lighter colors (pink) represent lower percentages. The main (

**top-left**) matrix contains numerical information, whereas the sub-matrices (

**bottom-left**,

**top-right**) contain percentages, also rearranged to present successful classifications percentages closer to the main matrix.

**Figure 7.**Confusion matrix for SqueezeNet. The colors in all matrices represent the values of the predictions. Darker colors (blue) indicate higher percentages, while lighter colors (orange) represent lower percentages. The main (

**top-left**) matrix contains numerical information, whereas the sub-matrices (

**bottom-left**,

**top-right**) contain percentages, also rearranged to present successful classifications percentages closer to the main matrix.

**Figure 8.**Training plot for SqueezeNet. The upper diagram illustrates the accuracy as it is formed after each training epoch, where the blue line represents the training accuracy, and the dark line represents the validation accuracy. The lower diagram depicts the loss, where the orange line represents the training loss and the dark dotted line represents the validation loss.

**Figure 9.**Depiction of the layers and architecture of the VGG-16 CNN used for classification. (

**a**) Architecture of the CNN containing all of the layers; (

**b**) magnified fragment demonstrating a partial view of the network layers and connections. The color scheme used is identical to the one detailed for SqueezeNet in Figure 6.

**Figure 10.**Training plot for VGG-16. The upper diagram illustrates the accuracy as it is formed after each training epoch, where the blue line represents the training accuracy, and the dark line represents the validation accuracy. The lower diagram depicts the loss, where the orange line represents the training loss and the dark dotted line represents the validation loss.

**Figure 11.**Confusion matrix for VGG-16. The colors in all matrices represent the values of the predictions. Darker colors (blue) indicate higher percentages, while lighter colors (orange) represent lower percentages. The main (

**top-left**) matrix contains numerical information, whereas the sub-matrices (

**bottom-left**,

**top-right**) contain percentages, also rearranged to present successful classifications percentages closer to the main matrix.

Actual Class | |||
---|---|---|---|

$\mathrm{YES}({P}_{e}$) | $\mathrm{NO}({N}_{e}^{i}$) | ||

Classifier’s Prediction | $\mathrm{YES}\left({P}_{c}\right)$ | TP | FP |

$\mathrm{NO}\left({N}_{c}\right)$ | FN | TN |

Accuracy | Specificity | Precision | Recall | F1 Measure | G-Mean |
---|---|---|---|---|---|

0.917 | 0.88 | 0.85 | 0.97 | 0.91 | 0.93 |

0.912 | 0.88 | 0.85 | 0.96 | 0.90 | 0.92 |

Accuracy | Specificity | Precision | Recall | F1 Measure | G-Mean |
---|---|---|---|---|---|

0.882 | 0.90 | 0.89 | 0.87 | 0.88 | 0.88 |

Classifier | Accuracy | Specificity | Precision | Recall | F1 Measure | G-Mean |
---|---|---|---|---|---|---|

MLP | 0.681 | 0.755 | 0.719 | 0.608 | 0.655 | 0.674 |

SVM | 0.799 | 0.865 | 0.846 | 0.733 | 0.784 | 0.796 |

KNN | 0.699 | 0.914 | 0.850 | 0.485 | 0.617 | 0.665 |

Naive Bayes | 0.736 | 0.798 | 0.770 | 0.674 | 0.718 | 0.733 |

Bag of Decision trees | 0.651 | 0.601 | 0.636 | 0.700 | 0.666 | 0.649 |

Random Forest | 0.913 | 0.938 | 0.934 | 0.889 | 0.910 | 0.912 |

AlexNet | 0.891 | 0.860 | 0.823 | 0.935 | 0.875 | 0.897 |

SqueezeNet | 0.917 | 0.878 | 0.846 | 0.974 | 0.905 | 0.925 |

VGG-16 | 0.882 | 0.898 | 0.886 | 0.866 | 0.876 | 0.882 |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Sideris, N.; Bardis, G.; Voulodimos, A.; Miaoulis, G.; Ghazanfarpour, D.
Enhancing Urban Data Analysis: Leveraging Graph-Based Convolutional Neural Networks for a Visual Semantic Decision Support System. *Sensors* **2024**, *24*, 1335.
https://doi.org/10.3390/s24041335

**AMA Style**

Sideris N, Bardis G, Voulodimos A, Miaoulis G, Ghazanfarpour D.
Enhancing Urban Data Analysis: Leveraging Graph-Based Convolutional Neural Networks for a Visual Semantic Decision Support System. *Sensors*. 2024; 24(4):1335.
https://doi.org/10.3390/s24041335

**Chicago/Turabian Style**

Sideris, Nikolaos, Georgios Bardis, Athanasios Voulodimos, Georgios Miaoulis, and Djamchid Ghazanfarpour.
2024. "Enhancing Urban Data Analysis: Leveraging Graph-Based Convolutional Neural Networks for a Visual Semantic Decision Support System" *Sensors* 24, no. 4: 1335.
https://doi.org/10.3390/s24041335