A Multimodal AI-Assisted Women Safety Prototype Using Random Forest, Opencv, Speech Recognition and GPS-Based SOS On Streamlit

This paper presents and critically evaluates a prototype women-safety application that integrates machine-learning-based risk prediction, computer-vision-based video analysis, voice emergency detection, GPSbased SOS messaging, and incident analytics in a single Streamlit dashboard. The prototype takes contextual safety information as input, including device type, panic-button status, response time, day, location category, age, and emergency-contact status, and passes it to a stored Random Forest classifier. Its computer-vision module uses OpenCV to measure inter-frame motion in uploaded video as an early-warning indicator, while a background listener detects predefined emergency phrases from microphone input. The supplied dataset contains 1,000 incident records; descriptive analysis shows a mean response time of 16.05 minutes, 504 panic-button incidents, and 482 assault records. The results demonstrate a functioning safety workflow, but the prototype cannot yet be considered a crime-detection system, because the project files contain no video annotations, training code, test metrics, or end-to-end threat classifier. The paper concludes with the validation, privacy, and deployment work required before field use.