Summary: | This study explores using different machine learning techniques and workflows to predict crime related statistics, specifically crime type in Philadelphia. We use crime location and time as main features, extract different features from the two features that our raw data has, and build models that would work with large number of class labels. We use different techniques to extract various features including combining unsupervised learning techniques and try to predict the crime type. Some of the models that we use are Support Vector Machines, Decision Trees, Random Forest, K-Nearest Neighbors. We report that the Random Forest as the best performing model to predict crime type with an error log loss of 2.3120.Comment: 11 pages, 55 figures, fixed typos, added references in Introduction sectio
|