A HYBRID DEEP LEARNING APPROACH FOR ONLINE RECRUITMENT FRAUD DETECTION USING BERT, ROBERTA, AND CNN2D

Abstract

With the increasing rise of digital employment platforms, online recruitment fraud (ORF) has become a critical cyber concern, where fraudsters use phony job advertising to take advantage of job searchers. An improved deep learning-based system that combines transformer models—BERT and RoBERTa—with sophisticated data balancing strategies that employ several SMOTE variants to efficiently handle class imbalance is proposed in this study to address this problem. A Convolutional Neural Network (CNN2D) is also added to the system for better feature extraction, which helps the model recognize intricate textual patterns and boost classification results. By merging job listings from several sources, a new and updated dataset is created, improving generalization and flexibility to current fraud trends. Metrics including accuracy, precision, recall, and F1-score are used to assess the models' performance after they have been trained and assessed on both original and balanced datasets. A Flask-based web application is also created to offer a user-friendly interface for real-time fraud detection, enabling users to enter job information and get predictions right away. When compared to conventional machine learning techniques, experimental findings show that the suggested hybrid framework greatly increases detection accuracy and resilience, making it a useful tool for spotting fake job listings.