This work focuses on the development of an advanced crime detection system using video surveillance data to enhance public safety. Detecting crime in video surveillance data presents several challenges, including the vast amount of data generated, the need for real-time analysis, and the complexity of distinguishing between normal and suspicious behavior in diverse environments. Traditional methods struggle to meet these demands due to their limited ability to process large datasets and accurately identify nuanced patterns. In this context, machine learning (ML) is crucial, as it enables the automation of complex pattern recognition and prediction tasks, making it possible to efficiently detect potential criminal activities in real time. By leveraging the capabilities of convolutional neural networks (CNNs) and long short-term memory (LSTM) networks, the system aims to detect and predict criminal activities in real time. The CNNs extract spatial features from video frames, while the LSTM networks capture temporal dependencies. The integration of CNN with LSTM shows significant improvements in crime detection accuracy, making it a robust prediction ML model for public safety applications. The hybrid CNN+LSTM model demonstrated a significant improvement over the CNN-only approach, achieving approximately 90% prediction accuracy on both the training and test data.