نوع مقاله : مقاله پژوهشی
نویسندگان
1 استاد، گروه مدیریت صنعتی، دانشکده مدیریت، دانشگاه تهران، تهران، ایران.
2 دانشجوی دکتری، گروه مدیریت صنعتی، دانشکده مدیریت، دانشگاه تهران، تهران، ایران.
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Introduction and Objectives: In today's In a highly competitive environment, recruitment decisions can no longer rely only on human judgment. The increasing volume of applicant data and the complexity of the search for candidate attributes and high precision in the selection of personnel have become a need. Artificial intelligence (AI) and machine learning have been chosen as the only solution to such a problem. (ML) a strategic necessity for organizations. Despite that, classical ML models, such as decision trees and logistic regression, are giving acceptable However, when results are applied to imbalanced datasets, complex data structure setups, and high accuracy requirements, they are greatly limited. This study aims to achieve a hybrid machine learning model designed on the forces of both kinds of neural networks as well as the classical algorithms. I demonstrate how to deliver a powerful, accurate, and interpretable solution to predict recruitment outcomes.
Methods: A multi layer stacking architecture was used to develop the proposed model, in which Deep Neural Network (DNN) is employed with four of the high performing base learners such as Random Forest, Gradient Boosting, LightGBM and CatBoost. Finally, XGBoost was used as meta learner to learn the final prediction from the outputs of these base models. To handle the class imbalance problem, NearMiss undersampling technique was tried and we used the Tree structured Parzen Estimator (TPE) algorithm provided as a part of the Optuna framework for hyperparameter optimization. Additionally, Recursive Feature Elimination with Cross Validation (RFECV) was used for feature selection to find the most important variables related to the hiring decisions.
Findings: The proposed hybrid model has been evaluated on a sample dataset of 1500 samples against 16 well-known machine learning models. Results indicated that the proposed model surpassed all key performance metrics in all areas of accuracy, precision, recall and F1 score with an accuracy of 92.47% and F1 score of 92.12%. There were some other models such as CatBoost and LightGBM that also had good scores, no other models performed better than those metrics reported for the proposed model.Likewise, the feature importance assessment of the same dataset with the help of XGBoost displayed that the recruitment strategy, education level, and interview score were the major predictors of final hiring decisions. These findings were not only beneficial in improving model performance but also valuable for improving the research and data examination of the HR decision makers in relation to the policies and criteria used in recruitment.
Conclusion: This research develops the hybrid machine learning model that smoothly combines classical algorithms and deep learning by a stacked architecture, which provides an advanced and highly effective structure for predicting hiring outcomes accurately. The model achieved both statistical superiority in benchmark comparisons and practical benefits.These findings imply that the usage of such hybrid models can rewrite the context for intelligent HR systems by streamlining candidate evaluation as faster, fairer, and more data-driven. In addition, HR managers receive focused, evidence-based feedback from feature analysis when predicting with modeling. Future work involving larger datasets and unstructured data such as resumes and interview videos coupled with tools for making the black box more explainable, such as SHAP or LIME, is encouraged to add transparency and build organizational trust in AI-based decision-making systems.
کلیدواژهها [English]
)