Intelligent ML Based Threat Detection System For Child Harassment and Online Predator Prevention

Abstract

The rapid growth of online communication platforms has increased the risk of child harassment and grooming by online predators. Research shows that such exploitation is typically a gradual process involving repeated interactions, subtle linguistic cues, and the sharing of digital content rather than explicit abuse. Traditional rule-based monitoring systems are ineffective in identifying these earlystage behaviours due to the large volume of unstructured data and evolving offender strategies. Recent studies highlight the effectiveness of machine learning and natural language processing techniques in detecting suspicious communication patterns through textual and behavioural analysis. This work presents an intelligent machine learning–based threat detection system designed to identify online child harassment and predator activities using multiple input formats, including manual text input, PDF and Word documents, JSON chat log files, and images. Textual data is processed using preprocessing and TF-IDF feature extraction, followed by Support Vector Machine classification to detect grooming and harassment patterns. Image inputs are analysed using convolutional neural network–based feature extraction to identify inappropriate visual content. By combining evidence from textual and visual sources, the system generates a reliable threat assessment. The summarized findings from existing research indicate that integrating linguistic, behavioural, and media-based analysis significantly improves detection accuracy while reducing false positives. The proposed approach offers a scalable and practical solution for early detection and prevention of online child exploitation, making it suitable for deployment in realworld digital safety environments.