An Integrated Framework for Automated Resume Screening Using RoBERTa, Random Forest and Explainable AI

The resume screening process is a critical stage in recruitment, yet conventional methods and traditional applicant tracking systems (ATS) often rely on manual review or keyword matching, resulting in slow, biased, and less objective evaluations. This study proposes an integrated automated screening system that combines RoBERTa for contextual feature extraction, Random Forest for candidate classification, and SHAP-based Explainable AI for interpretable decisions, enhancing transparency, efficiency, and fairness beyond traditional ATS. The dataset consists of real resumes and synthetically generated ones designed to mimic the distribution of real data, with K-means clustering used to establish labeling thresholds. Experimental results show that RoBERTa achieved an F1 Score of 81.08% in feature extraction, while Random Forest reached 96% accuracy in suitability classification. SHAP-based explanations provide insights into feature contributions for each prediction, offering an actionable understanding for recruiters. This integrated framework not only improves the efficiency and fairness of resume screening but also demonstrates a practical application of explainable AI in recruitment.