Identification of Crime using Multi Embedding BiLSTM

Abstract

Crimes pose significant societal challenges with implications for a nation's well-being, economic progress, and reputation. Precisely measuring crime rates, categories, and hotspots from historical patterns presents various computational complexities and opportunities. This study introduces and improves a deep learning approach for predicting crime types with high precision. The system can predict both crime categories and associated risk levels by analyzing concise summaries from criminal case reports. The predictive model is built on a neural network with LSTM and Bi-LSTM components, demonstrating remarkable accuracy in forecasting crime types despite limited training attributes. It is tested on a substantial real-world dataset containing historical urban crime data, offering a deep learning-based solution to enhance public safety in the face of criminal activities.