A Model for Predicting Crime Risk

Efficient resource allocation and effective risk management are paramount for governments and police forces. This article presents a comprehensive model designed to predict crime risk levels, with the goal of optimising resource allocation and reducing the associated time, cost, and effort in risk management. By considering crucial factors such as location, time, and crime type, our model endeavours to provide accurate and actionable insights. To ensure the reliability of our predictions, we implemented various data wrangling techniques, including feature selection, data validation, and the creation of new measures. These steps were instrumental in preparing the data for analysis and generating reasonably accurate results. Additionally, we explored a range of machine learning algorithms, namely Logistic Regression, Gaussian Naive Bayes, Decision Tree, XG Boost, and Random Forest, to predict crime risk levels. The models were meticulously validated using cross-validation techniques and evaluated based on diverse performance metrics. In a bid to further advance our predictive capabilities, we leveraged deep learning techniques with TensorFlow, enabling a performance comparison against traditional machine learning models. Notably, the Random Forest algorithm has emerged as the most effective, yielding an impressive accuracy of 90%. The culmination of our efforts is a successful software application, complete with a user interface integrated with our cutting-edge prediction model.