Detection and Subclassification of Severity Levels of Crime Activities with Machine Learning

Crime can be described as the intentional commission of an action that is often labelled as harmful to society, prohibited, and punishable under criminal law. The increase of criminal activities upon abolishment of the COVID-19 lockdown has been imminent as seen in a handful of first world western countries, whereby mass riots and lootings occur ever so frequently. The detection of the specific crime activities corresponding to the severity levels of the crime is therefore important which could help timely dispatch and management of government officers to the crime scene. The classification was conducted using various machine learning models such as Convolutional Neural Network (CNN), 3D CNN, Long Short-Term Network (LSTM), and transfer learning of Densenet and VGG-16 architecture. Based on the findings, 3D CNN recorded the highest performance above 95% for all the assessed metrics which significantly outperformed other conventional machine learning models as mentioned earlier. The results are mainly limited by the availability of dataset as to date dataset related to various severity levels of crime activities are limited and continuing development in this field is necessitated.