"I quit because…": A psychological interpretation of push-pull dynamics of employee attrition informed by explainable...

Employee attrition is a significant concern in today's organizations because it affects productivity, workforce stability, and long-term talent retention. This study aims to develop and validate a Precision HR Diagnostic Framework to identify and interpret the key drivers of employee turnover at both organizational and individual levels. This study uses the IBM Employee Attrition dataset, consisting of 1470 employee records and 35 features on employee demographics, job satisfaction, and performance metrics, and evaluates eight ML algorithms to benchmark our proposed framework. It utilizes Shapley additive explanations (SHAP) and Permutation feature importance (PFI) based explainable Artificial Intelligence (AI) techniques to interpret the ML results. The empirical results show that over time, job satisfaction, environmental satisfaction, stock options, salary, and work-life balance are the key predictors of attrition. It identifies AdaBoost as the top model based on F1 score, accuracy, and interpretability characteristics. Interpreting these results through the Self-Determination Theory (SDT) and Job Demands-Resources (JD-R) frameworks reveals a distinct push-pull mechanism: job demands, such as overtime and work pressure, act as push factors driving turnover, while resources, such as higher salaries and stock options, act as pull factors leading to employee retention. Unlike prior studies that primarily focus on improving predictive accuracy, this work develops a statistically rigorous, explainable machine learning framework that provides a psychologically grounded, theory-driven approach that facilitates employee retention by strategically balancing pull resources against unavoidable push factors using SHAP.