Automated resume screening using machine learning: an ensemble-based approach for efficient candidate selection

The study introduces an innovative Machine Learning (ML) based ensemble framework to automate the process of resume screening, aiming to increase recruitment efficiency. The research employs a two-phase experimental design: initially, several standalone ML classifiers such as Logistic Regression, Decision Trees, Naive Bayes, and Random Forest are evaluated to establish baseline performance.

Subsequently, the most effective models are combined in a strategic manner using a soft-voting mechanism to create an ensemble framework. This ensemble model demonstrated superior accuracy compared to individual classifiers, achieving an impressive 98.44% accuracy (5-fold cross-validation mean: 97.47% ±1.00%).

To ensure the validity of the results, paired T-tests (with a significance level of p < 0.05) were conducted, confirming the reliability of the performance gains. The framework's robustness was further tested on a new, curated dataset comprising 598 expert-labeled records, all labeled in a zero-shot testing protocol. The ensemble framework exhibited remarkable stability, with only a minor 1.10% decrease in performance and an F1-score of 0.99, underscoring its reliability in diverse scenarios.

The research also focused on ensuring the fairness and ethical application of the algorithm. An interpretability audit was conducted to verify that the selection process remains merit-based, thereby addressing concerns related to algorithmic bias. The study culminated in the development of a fully functional web-based prototype that showcases the practical viability of the framework in real-time resume classification and recommendation, making it highly suitable for real-world talent acquisition.

The research was supported by the Deanship of Scientific Research and Libraries at Princess Nourah bint Abdulrahman University, through the "Nafea" Program, Grant No. (NP-45-089). The findings are published under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, allowing non-commercial use and sharing of the research as long as proper attribution and citation are provided.