A Hybrid Artificial Intelligence Framework for EmotionAware Domestic Violence Detection Using Video Analytics

Abstract

The increasing use of digital communication platforms has created opportunities for identifying emotional behavior and detecting potentially harmful activities through intelligent data analysis. This paper presents an artificial intelligence–based framework for emotion recognition from social media text with an extension for identifying violent activities from surveillance videos and facial expressions captured through a live webcam. The proposed system utilizes a Twitterbased emotion dataset containing positive, neutral, and negative sentiments. Initially, the textual data undergoes preprocessing, including noise removal, stop-word elimination, tokenization, and feature extraction using the Term Frequency–Inverse Document Frequency (TF–IDF) technique. The processed data are divided into training and testing sets using an 80:20 ratio. Multiple machine learning and deep learning models, including Naïve Bayes, Decision Tree, Random Forest, Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM), are trained and evaluated to analyze their emotion classification performance. Experimental results indicate that ensemble and deep learning approaches achieve superior classification accuracy while maintaining reliable prediction capability across different emotional categories. The developed framework further extends its functionality by examining CCTV video streams to recognize violent activities and by employing a webcam-based facial emotion recognition module for real-time monitoring. By integrating text analytics, computer vision, and deep learning techniques into a unified framework, the proposed system demonstrates an effective approach for supporting early identification of emotional distress and suspicious activities, offering potential applications in public safety, digital surveillance, and intelligent monitoring systems.