Deep Neural Network-Based Approach to Identify the Crime Related Twitter Posts

Crime prediction is an important task to reduce criminal activities in the society. We can minimize the harm by identifying the crimes hotspot before any crime happens using a crime prediction model. Thus, crime prediction is becoming a hot topic among researches. We can identify crime patterns by analyzing historical crime data and predict future crimes. Researchers used different sources to get crimes related data to generate the prediction model. But, some crimes are unregistered. In this paper, we used twitter posts to detect crimes. People share information around their environment via Twitter posts. Here, we proposed the Bidirectional Encoder Representations from Transformers (BERT) approach to identify the crime-related posts. Our approach outperformed the existing approaches by obtaining 92.8% accuracy.