Crime Prediction in Swedish Municipalities with machine learning algorithms

Abstract: In this thesis we use a number of common machine learning algorithms to predict crime rates in Swedish municipalities. As predictors we use a mix of municipal socioeconomic variables. For some years we are able to correctly classify up to 85% of the municipalities that have a high crime rate. The highest prediction accuracy rates are obtained from tree and clustering based models. Important factors for forecasting crime in Sweden seem to be divorce rates, male age, unemployment and unsuccessful high school education.