Authors: Shraddha Rajesh Baisane, Professor Bangar A.P., Professor Bhosale S.B.
Abstract: Women's safety is still an everyday concern in public places, workplaces and homes, and most tools in use today de-pend on the victim pressing a button at the worst possible moment. Artificial intelligence (AI) offers another route: lis-tening for distress in the surroundings, combining it with other signals, and forecasting where and when the risk is high-est. This paper reviews 15 IEEE-published sources from 2014 to 2024 that relate to this idea. They cover IoT-based safety devices, scream and distress detection from audio, speech emotion and fear recognition, multimodal wearable systems, crime hotspot prediction, and edge and privacy-preserving deployment. The review finds that the audio models report accuracies between 92% and 100% on the datasets they were tested on, while the one multimodal wearable study that tested fear recognition on unseen subjects reached 63.61%. Crime-prediction methods are mature at city level but are rarely linked to women-specific incidents or to live sensing. Most studies rely on small, public or acted data, say little about false alarms, response delay and battery use, and give limited attention to privacy, accents and noisy streets. The paper contributes a comparison of eleven representative works, a conceptual seven-layer architecture that joins on-device audio detection with predictive risk analytics, and a prioritised list of research gaps. The architecture is conceptu-al and has not been tested experimentally.
