Published Paper Details:

Criminal activity is consistently one of the most significant societal issues, endangering both individuals and public safety. The government, police, and public may all implement efficient crime prevention strategies with the aid of accurate crime prediction. This study reviews the literature on crime prediction methodically from several temporal and spatial angles. With an emphasis on prediction techniques, we provide an overview of the state of crime prediction research as of right now from four angles: prediction content, crime kinds, methodology, and assessment. Crime prediction on different temporal and spatial scales may be broken down into three categories: micro-, meso-, and macro-level prediction for spatial crime, and short-, medium-, and long-term prediction for temporal crime prediction. A range of assessment criteria and crime prediction techniques are also compiled, and various models and prediction techniques are contrasted and assessed. After reviewing the literature, it was discovered that there are still a lot of gaps in the knowledge base. These gaps include: (i) the difficulty of effectively handling data sparsity; (ii) the lack of predictive model practicality, interpretability, and transparency; (iii) the evaluation system's relative simplicity; and (iv) the paucity of research on the application of decision-making. To address the issues mentioned above, the following recommendations are made in this regard: In order to deal with sparse data, (i) transformer learning technology is used; (ii) model interpretation techniques, such as Shapley additive explanations (SHAPs), are introduced; (iii) a set of standard evaluation systems for crime prediction at various scales is established in order to standardise data use and evaluation metrics; and (iv) reinforcement learning is integrated in order to achieve more accurate prediction while encouraging the transformation of the application results.