Smart Police: A Hybrid Deep Learning Model for Crime Proactivity Assessment

Authors:

Abstract
Conducting competent and precise investigations on criminal situations is necessary to establish peace, order, and justice in a community. With the increased use of modern sophisticated systems to track crime and violence, computer programmers may also significantly assist officials in law enforcement. Previous research was not successfully able to accomplish crime forecasting and precision of predictions made by learning models. This paper proposes CNN-LSTM hybrid approach for forecasting and predicting criminal behavior with improved accuracy. To understand the crimes data, the study uses different machine learning models such as Support Vector Machines, Naïve Bayes, k-Nearest Neighbors, Logistic Regression, Multilayer Perceptron’s (MLP), cloud gradient boosting (XGBoost), decision tree, random forests, time series analysis implementing a Convolutional Neural Network and (ARIMA) Autoregressive Integrated Moving Average and Long-Term Memory model and machine learning algorithm to improve error predictions. The demonstration of model was evaluated with various performance criteria. After extensive testing on real-time data, the CNN-LSTM model shows higher accuracy compared to existing models. Specifically, the model achieved 91% accuracy in the dataset of Denver and 90% in the Los Angeles dataset.

This publication has 25 references indexed in Scilit: