Cognizable crime rate prediction and analysis under Indian penal code using deep learning with novel optimization approach

The exacerbation of high crime rate has become a critical impediment to the country's economy, therefore necessitating the involvement of data analysts and scientists in extracting invaluable insights from crime data to proffer solutions towards crime prevention and regulation. To this end, this study presents a novel Particle Swarm-Cuckoo Search (PS-CS) optimization algorithm which leverages the potency of both particle swarm optimization and cuckoo search algorithms to enhance the optimization of network parameters and enable the training of deep neural networks for crime prediction. The proposed PS-CS model supersedes the traditional backpropagation algorithm which is fraught with limitations such as slow convergence and a proclivity for local optima. The implementation of the proposed model has the potential to be an efficacious tool for predicting crime rates in India, thus facilitating the efforts of law enforcement agencies in controlling and curbing criminal activities. We conducted a comprehensive and intricate evaluation of the efficacy of our proposed methodologies vis-a-vis the most pervasive and prevalent classification models currently in use. These models encompassed not only conventional machine learning techniques, such as Multiple Linear Regression, Support Vector Regression (SVR), Random Forest Regression, and Decision Trees, but also advanced and state-of-the-art deep learning models such as Residual Neural Network-152 (ResNet), Visual Geometry Group (VGG), and EfficienteNet-B7. Our findings evince that the suggested approach, PS-CS, in conjunction with CNN model, outperformed all other models, yielding an unparalleled accuracy score of 99.87%. By comparison, other models, including Multiple Linear Regression, SVR, Random Forest Regression, and Decision Trees, exhibited markedly inferior accuracy scores ranging from 80% to 95%. It is therefore evident that our proposed methodology, when integrated with a deep learning model, proves to be an exceedingly efficacious and robust solution for classification tasks.