Enhancing Crime Predictions with Graph Machine Learning: A Survey on Techniques for Inconsistent Data

Graph machine learning (ML) techniques have gained a lot of attention lately and have advanced significantly in graph applications. The majority of graph machine-learning techniques have so far only been tested on social networks; the field of criminal informatics has not yet seen a thorough analysis of these techniques. Computationally, accurate crime prediction and future trend forecasting can help to improve urban safety. The inability of humans to comprehend complicated information from large amounts of data makes it difficult to anticipate and predict crime in an early and accurate manner. Numerous computational opportunities and challenges arise from precisely calculating the crime rate, types, and hot locations based on historical patterns. This paper presents a review of graph machine learning techniques and their uses in the field of crime prediction using data from electronic crimes at two separate levels: Link prediction and node classification. For these two layers, graph neural networks (GNN) and shallow embedding are popular graph machine-learning techniques. We examine the current state and emerging trends in graph machine learning techniques for predicting crime using electronic crime data based on the studies found. Although there are remarkable outcomes of GNN-based models as compared to conventional machine learning techniques in making predictions, there are some issues related to the interpretability of dynamic graphs. While machine learning techniques for crime prediction are still in their infancy, GNN-based models show a promising method for crime prediction.