Social media based crime prediction using machine learning

Crime prevention relies on utilizing crime prediction as a pivotal method to determine the most efficient patrol strategy for law enforcement agencies. Various methods and solutions have been employed to forecast criminal activities. However, the landscape and data involved in crime prediction are constantly evolving. Despite its potential as a valuable source for gathering insights, social media content has often been overlooked by predictive models. Therefore, this study aims to investigate the possibilities and benefits of integrating sentiment analysis and crime data using an innovative machine learning model to predict crime. To achieve this goal, the study proposes an approach based on ConvBiLSTM (Convolutional Bidirectional Long Short-Term Memory) for both sentiment analysis and crime prediction. This approach combines a CNN and Bi-LSTM. The CNN efficiently extracts high-level local features, while the BiLSTM enhances the contextual information available to deep learning algorithms. We also apply this approach to sentiment analysis of tweets and fake news detection to verify its effectiveness. Furthermore, the research employs feature-level data fusion to integrate information from both tweets and crime-related data. The study demonstrates several key findings (1) this approach achieves state-of-the-art performance in crime prediction using tweets, (2) it introduces a multimodal ConvBiLSTM model capable of integrating information from both crime and tweet data, (3) the study evaluates its performance against other baseline and benchmark models, establishing its superiority.