Scholarlink

<p>Employee attrition poses a significant challenge for organizations, resulting in substantial financial losses and operational disruptions. This paper presents a comprehensive approach to predict and explain employee attrition using normalized mutual information for feature selection and artificial neural networks (ANN) for prediction. We address the critical gap between model performance and interpretability by employing multiple explainable AI (XAI) techniques. Using the IBM HR Analytics dataset, we first apply normalized mutual information to identify and remove three redundant features with high correlation values, reducing multicollinearity while preserving predictive power. We then develop an ANN with dropout regularization with a final test accuracy of 82.7% and weighted averages of precision, recall, and F1-score of 0.85, 0.83, and 0.84, respectively. To enhance model transparency, we implement a multi-faceted explainability framework incorporating permutation feature importance, SHAP (originated from game theory), partial dependence plots, and accumulated local effects. Our analysis reveals that overtime status, environment satisfaction, job level, and stock option level are the most influential predictors of attrition. The XAI methods employed uncover complex interactions between features, such as the compound effect of distance from home and environment satisfaction on retention. Additionally, we provide a highly interpretable decision tree model as a transparent alternative. This comprehensive approach enables human resource professionals to not only predict at-risk employees but also understand the underlying factors driving attrition, facilitating targeted retention strategies and data-driven decision-making.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Feature selection and interpretation of employee attrition using mutual information and artificial neural networks

  • 期刊论文

  • 发表日期：

    2025年7月19日

查看全文
  • Khayyam Salehi,
  • Marziyeh Aghabalasafar

摘要

Employee attrition poses a significant challenge for organizations, resulting in substantial financial losses and operational disruptions. This paper presents a comprehensive approach to predict and explain employee attrition using normalized mutual information for feature selection and artificial neural networks (ANN) for prediction. We address the critical gap between model performance and interpretability by employing multiple explainable AI (XAI) techniques. Using the IBM HR Analytics dataset, we first apply normalized mutual information to identify and remove three redundant features with high correlation values, reducing multicollinearity while preserving predictive power. We then develop an ANN with dropout regularization with a final test accuracy of 82.7% and weighted averages of precision, recall, and F1-score of 0.85, 0.83, and 0.84, respectively. To enhance model transparency, we implement a multi-faceted explainability framework incorporating permutation feature importance, SHAP (originated from game theory), partial dependence plots, and accumulated local effects. Our analysis reveals that overtime status, environment satisfaction, job level, and stock option level are the most influential predictors of attrition. The XAI methods employed uncover complex interactions between features, such as the compound effect of distance from home and environment satisfaction on retention. Additionally, we provide a highly interpretable decision tree model as a transparent alternative. This comprehensive approach enables human resource professionals to not only predict at-risk employees but also understand the underlying factors driving attrition, facilitating targeted retention strategies and data-driven decision-making.