Predicting Crime Rate in Toronto

Predicting rates of criminal activity is a critical domain of study that has captured substantial interest in contemporary times. Various statistical, deep learning as well as machine learning techniques are being used to predict occurrence of crimes based on different factors such as time, location, and socioeconomic conditions. The use of big data analytics and predictive models are capable of assisting law agencies by taking premptive actions to avoid crimes and improve the safety of public. Several studies have shown that predictive models based on algorithms of machine learning like Decision tree, Neural Networks, and Random forest can provide accurate crime rate predictions. The efficiency of such models is mainly reliant on the quality of the information and the features used in analysis. In this project, A Random Forest classifier is used to create a classification model to predict the kind of major crime that has taken place based on various factors along with neighbourhood, division, time of crime, month, year and so on. In Toronto between 2014-2019, a detailed information including where and when the crime is committed and various other factors are included in our dataset, containing categorical data. The process involves in testing the data using both OneHot and numeric encoding. Despite the uneven distribution of instances, where assault records are in the majority, the model delivers satisfactory results on F1-score when dealing with the five class classification problem.Model’s performance doesn’t get affected while balancing the class weights. The model that uses Random Forest Classifier and One Hot Encoder shows a slight increase in accuracy, as indicated by an F1-score.