A Novel Machine Learning-Based Approach to City Crime Sensor Placement Prediction

The integration of advanced technologies can be up-and-coming for human-centered benefit. As the quality of life can be heavily affected by the safety standards of vehicular road traffic, the placement of appropriate traffic sensors can contribute to these standards. Collecting sensory traffic data involving location, speed, travel time, headways, and others is vital in designing traffic-resolving procedures. With such data, machine learning and data analysis tools are required to detect patterns and perform predictions. Through our research, we aim to investigate the potential of sensor data in predicting traffic crime hot spots using Machine Learning (ML) techniques, specifically, Random Forest, Long Short-Term Memory Network (LSTM), and Fourier Series Neural Networks (FSNN). Overall, we use three data sets, one with location codes, one with ticket outputs, and one demonstrating general crime rates in different locations within the City of Toronto. We run the models aiming for very low mean score error and the highest R2 score possible. Upon satisfaction with the initial evaluation, we run the model with only the unused portion of the crime rates data set to determine new traffic areas of interest for the sensors. This helps us develop sensory information to serve as traffic crime indicators and predictors.