Leveraging Generative Models, Deep Learning Architectures, and Machine Learning Classifiers for a Data-Driven Approach to Crime Detection and Prediction

Crime poses a significant challenge to the prosperity and growth of nations. Various factors, including poverty, deficiencies in the legal system, unstable economic conditions, and insufficient technological capabilities for crime analysis and detection, contribute to its emergence. This work proposes a novel framework for crime identification and detection utilizing generative and deep learning models. Initially, we extract latent features from an available dataset using a combination of generative model, Deep Learning (DL) and Machine Learning (ML), techniques, including a Variational Auto-Encoder (VAE), transformers, Convolutional Neural Networks (CNN), and Principal Component Analysis (PCA), followed by K-means clustering. We then evaluate the effectiveness of this clustering approach using renowned classifiers, such as Random Forest (RF), Decision Tree (DT), Gradient Boosting (GB), and Naive Bayes (NB). As a result, the framework, which utilizes VAE for feature extraction and combines it with RF as the classifier, achieves the highest accuracy of 0.9993. The strength of the proposed framework lies in its unsupervised learning approach, which attains significant information from data without relying on labeled datasets. This data-driven methodology adeptly leverages generative and deep learning models for feature extraction, subsequently employing these features for crime detection. Furthermore, we analyze individual attributes in latent spaces and apply classifiers, with the VAE demonstrating exciting performance, achieving an accuracy of 0.999.