A Proposed Approach for Crime Type Prediction using Machine Learning Techniques

Abstract— Crime and suspect prediction represent an immerging field where machine learning can be useful when applied efficiently. The available data can be used to train the proposed prediction model (building the model, classifier for example) and then this model can be tested and used to predict the crime type and suspect information such as sex, race and age category. The work in this paper presents a proposed approach for such prediction by using real world dataset available on the internet. The gathered data were preprocessed(deletion of the rows with missing and unknown attribute content, converting some attribute values into nominal or categorical data and concept hierarchy techniques) and reduced the number of attributes by choosing the most important features using different approaches and algorithms, then different Decision Tree types, Naïve Bayes Classifier and Association Rules prediction techniques were used to create the required models and to find out associations between different attributes of the given data. Testing phase shows high efficiency and effectiveness of the proposed approaches which in turn provide models that can be used as reliable predictors. An accuracy of more than 85% was achieved when using different classification techniques. WEKA software, Microsoft Excel and XLSTAT mining software were used to analyze the given data.