Intelligent Decision Support System for Crime Investigation

Abstract

Crime presents one of society's most formidable challenges, with its escalating rates significantly impacting individuals, families, and communities in detrimental ways. Therefore, efforts aimed at reducing the spread of crimes seek to reduce the spread of crimes by analyzing this large amount of data related to crime and accurately predicting it in real time. Thus, the competent authorities can take the necessary measures to prevent the occurrence of crime. Hence, advanced systems and contemporary methodologies for crime analysis are imperative to safeguard society. This paper presents an intelligent decision-support system in investigating crimes, using intensive crime data analysis to predict the quantity and types of crimes that occur at specific times and locations. Herein, Comparative analyzes are performed between machine learning and deep learning models, including Random Forest, K-NN, Naïve Bayes, ANN, Linear Regression, ARIMA, and LSTM, across datasets from the Urban Institute, Chicago, and Boston. The results indicate that linear regression outperformed the other methods in terms of regression, achieving higher R-squared values, particularly 0.946 and 0.476 for Chicago and the Urban Institute, respectively. On the other hand, LSTM exhibited in Boston by 0.525. For classification, Random Forest showed superior performance, with accuracies of 99.6%, 72.5%, and 78.6% for the Chicago, Boston, and Urban Institute datasets, respectively.