Online Recruitment Fraud Detection Using Deep Learning Approaches

In today’s digital landscape, many businesses are utilizing digital platforms to their advantage for recruiting new employees, streamlining the hiring process. However, the surge in online job postings has also led to an increase in fraudulent advertisements, with scammers profiting from deceptive job listings. This rise in online recruitment fraud has become a significant concern within the realm of cybercrime, making it essential to identify and eliminate fake job postings to protect job seekers. Recent research has explored the application of conventional machine learning and deep learning techniques techniques for detecting fraudulent job listings. This study aims to employ two transformer-based deep learning models and the Robustly Optimized BERT-Pretraining Approach (RoBERTa) to improve predictive precision of fake job detection. To support this research, a novel dataset of fraudulent job postings has been created by aggregating data from three distinct sources. Existing benchmark datasets are often outdated and limited in scope, which hampers the effectiveness of current models in identifying fraudulent job listings. Therefore, this study updates the dataset with the latest job postings. Exploratory Data Analysis (EDA) reveals a class imbalance issue in the detection of fake jobs, which can lead models to be overly aggressive towards the minority class. To overcome this challenge, this research utilizes ten advanced SMOTE variants to handle class imbalance. The resulting model performance is then assessed using standard evaluation metrics. models, adjusted by each SMOTE variant, is analyzed and compared. All approaches demonstrate competitive results, with the BERT model combined with the SMOBD SMOTE variant achieving the highest balanced accuracy and recall, reaching approximately 90%.