Fall 2016 class project for cs5984
Predictive policing is a term given to the practice of policy making using predictive, analytical and computer science techniques in the field of law and order to predict possible crimes and taking actions to prevent it. In this project, we apply data science techniques to characterize and predict crime in the city of Chicago. We aim to develop a model that predicts crime for a given location and on a future date, when information such as the presence of nearby restaurants and police stations, historical weather and historical crime data is known. We first perform a qualitative analysis of crimes that were reported in the year 2006, and later present a binary classification problem that determines whether a particular area is susceptible to a crime event or not. We develop this model using different supervised machine learning algorithms like k-Nearest Neighbor, Decision Tree, Logistic Regression and Random Forest. We evaluate each of these method under training data-sets of different size. \textbf{Our model evaluation shows that on average, K-NN performs the best, predicting with an f1-score of 0.92 but Random Forest classifier predicts with the highest recall of 0.99}.