Local Interpretable Model-agnostic Explanations for Crime Prediction

Crime is one of the most significant societal challenges, that poses a serious threat to public safety and individuals. Predicting crime accurately can assist governments, law enforcement, and communities in implementing effective crime prevention strategies. For this reason, recently, machine learning approaches have been applied to identify patterns and trends in crime occurrences and, as a consequence, predict crimes. However, these techniques, despite their good performance, have the drawback of producing incomprehensible decisions, which could limit their use in the real context. To address this limitation, this paper introduces the use of an explainable artificial intelligence technique, namely Local Interpretable Model-agnostic Explanations (LIME), to provide explanations for crime predictions generated by neural networks. The experimental results demonstrate that LIME is a good approach to understand the rationale behind machine learning-based crime prediction.