Crime Prediction for Stop and Search Outcomes using Machine Learning
A stop and search are a preventive act that used by the law enforcements to predict crimes and prevent them from just relaying on the experience of the officers to decide whom to stop and search. In this research, machine learning is used to predict the outcomes of the stop and search using data from UK police for London city from 2015 to 2017. This research compares nine difference machine learning classifiers for predicting multiple outcomes of the result of the search events. The study proposes feature grouping for the self-stopped ethnicities as well as classes grouping. The result of the experiments during all five stages CNN was among the top three classifiers and 16 out of 20 evaluation metrics followed by the LR that was among the top three and was the best for the last two stages of the experiment. Our proposed feature grouping approach (i.e. grouping the ethnicities and outcomes) has really performed well on all classifiers and specifically on CNN where the accuracy was 85% and 0.8 recallM 75% F-measureM and precisionM.
Index Terms - Crime predication, Machine learning, law enforcements, Artificial intelligence systems