人工知能学会

Employee attrition, the workforce reduction in organizations, is traditionally viewed negatively in human resource management literature, causing disruptive changes. Limited access to sensitive employee data complicates analysis. This study introduces a comprehensive framework, involving data cleaning, feature extraction, and dataset normalization through exploratory data analysis (EDA), encompassing univariate and bivariate analysis. Utilizing Kaggle HR Analytics and IBM HR Analytics datasets, we tackle challenges associated with imbalanced data. To address dimensionality issues, various feature selection techniques are incorporated. Attrition prediction employs machine learning classifiers—Logistic Regression, Random Forest, MLP, Decision Tree, AdaBoost, and Boost. SMOTE is applied to counter class imbalance. Our approach utilizes machine and ensemble learning on both large and normal-sized HR datasets, achieving state-of-the-art performance in accuracy and AUC scores. The study's segmentation technique provides HR managers with diverse groupings of employee attributes, offering valuable insights for developing effective retention strategies.