EMOTION RECOGNITION WITH MACHINE LEARNING FOR WOMEN'S SAFETY

Women's safety in public environments is a global problem, exacerbated by high rates of violence and vulnerability. A critical challenge is to identify emotional signals accurately, given the variability of facial expressions in risky situations. This work aims to develop an emotion recognition system to prevent and respond to harassment, differentiating signs of fear and sadness in women and anger in men. The FER-2013 and RAF-DB databases were used, with the MobileNetV2 Convolutional Neural Network and transfer learning on the Edge Impulse platform. Automated processing and labeling ensured data consistency. The experiments obtained overall accuracies of 85.7% and 86.9%, and the accuracy metrics ranged from 79.57% to 90.24%, proving the model's performance. In the future, the system could be integrated with surveillance devices, expanding the response to harassment.