DEVELOPMENT OF A PREDICTIVE MODEL FOR DETECTING FAKE JOB POSTINGS USING FEATURE EXTRACTION AND DEEP LEARNING.

The rise of fake job postings on platforms such as LinkedIn and Indeed has become a serious issue, exposing job seekers to risks including identity theft, financial loss, and emotional distress. Traditional detection methods based on keyword filtering or shallow machine learning have proven insufficient in identifying the deeper contextual patterns of fraudulent listings. This study developed a deep learning-based predictive model to accurately classify job postings as genuine or fake by combining structured features (e.g., job title, location, employment type) with unstructured textual data (e.g., job descriptions and company profiles). Using a dataset of about 18,000 postings from Kaggle, preprocessing steps included handling missing values, applying NLP techniques such as tokenization, stop-word removal, and TF-IDF transformation, and encoding structured variables. A TF-IDF model was trained and evaluated across three data splits (60/40, 70/30, 80/20) using metrics like accuracy, precision, recall, F1-score, and AUC. The best performance was achieved with the 80/20 split, yielding 98.21% accuracy, 92.1% precision, 70.7% recall, and an AUC of 0.852, indicating strong classification capability. However, overfitting was consistently observed, limiting generalization to new data. The study recommends strategies such as early stopping, SMOTE for class imbalance, and threshold optimization to improve reliability in real-world deployment.