Advanced Ensemble Way Automated Resume Screening Using Dual-Level Features and Blockchain

Abstract

This paper introduces an advanced ensemble model for automated resume screening that achieves the best performance through dual-level text representation and weighted classifier fusion. We propose a novel approach combining character n-grams (3–5) and word n-grams (1–2), along with an ensemble of Support Vector Machine, Logistic Regression, and Gradient Boosting classifiers (optimised prob-based voting). Evaluated on 24 categories of jobs with 2484 resumes, our proposed model achieves 67.7% accuracy and 0.665 F1-score, which is significantly better than the baseline methods (gradient boosting: 63.8%, logistic regression: 59.1%,character-level SVM: 53.0%). We further develop a blockchain-based audit trail to record the resume identifier's hash, the model version, and decisions, creating unchangeable, verifiable logs to support accountability and compliance. Fairness analysis employing synthetic attributes and exhaustive ablation is a study that validates the effectiveness of our approach. Index Terms–Resume screening, recruitment, fairness in AI, convolutional neural network, transformers, blockchain, auditability.