SafeHer: A Proactive AI-Driven Safety System for Women Using Contextual Risk Assessment and Real-Time Monitoring

Abstract: Women’s safety remains a critical global challenge, particularly in urban and semi-urban environments where conventional manual panic-button applications fail precisely when they are needed most. This paper presents SafeHer, a proactive, AI-driven mobile safety platform that shifts personal protection from reactive alerting to predictive, autonomous risk assessment. The system continuously monitors a user’s contextual environment—GPS location, motion patterns, time-of-day, network signal strength, and environmental isolation—and processes these signals through a two- stage Hybrid Danger Score Engine. The first stage applies deterministic rule-based thresholds on-device (Phase 1), producing a Base Risk Score (0–50) within milliseconds and without network dependency. The second stage transmits a structured context vector to Google Gemini AI via a secure Firebase Cloud Function proxy, obtaining an AI Risk Score (0–50) with contextual reasoning and one-sentence justification. The combined Final Danger Score (0–100) drives a threshold-triggered autonomous alerting pipeline that dispatches Firebase Cloud Messaging push notifications and Twilio SMS to pre-registered trusted contacts, shares a live GPS tracking link, and logs incident context—all without requiring any user interaction. Beyond automated alerting, SafeHer includes a Fake Call module for discreet social exit, an Evidence Vault with a system-update decoy screen for covert audio recording, a community Fear Map with differential privacy (Laplace noise, ε = 0.1) for anonymized urban safety analytics, and a Trust Score system for crowd-validating community hazard reports. Built on a Flutter frontend with a scalable Firebase backend, the platform achieves average alert latency of 1.8–2.7 seconds, background battery consumption of 9.3% over eight hours, a hybrid AUC of 0.94, and a System Usability Scale score of 82.4 across 30 simulated threat scenarios and a 15-participant pilot study.

Keywords: Women Safety, Predictive Safety Systems, AI-Based Risk Detection, Mobile Context Sensing, Firebase Architecture, Google Gemini AI, Proactive Alerting, Real-Time Location Tracking, Fake Call Feature, Evidence Vault, Fear Map, Trust Score, Differential Privacy.