Investigating employee attrition using machine learning techniques

Introduction: This study investigates underlying issues that employees might not openly disclose in exit interviews by leveraging machine learning techniques to explore the factors causing employee turnover, offering insights beyond churn predictions and traditional exit interviews. The novelty of this research lies in the use of ML causal inference to draw conclusions.

Methods: The machine learning algorithm was trained on 10 features of the dataset with 14,999 records. The feature importance analysis and clustering highlighted the most influential factors in predicting attrition. Then, propensity score matching was used to estimate the causal effect of these features on attrition by comparing similar groups of employees who stayed and left. 

Results: The model achieved an impressive accuracy of 95.25% and an F1-score of 96.0%, demonstrating the robustness of the algorithm. Further analysis, including clustering and causal inference using propensity score matching, revealed distinct patterns among departing employees, such as low, frustrated, and high performers.

Conclusion: By employing causal inference rather than merely prediction, this study offers a more objective understanding of the causes of attrition. The causal model in this research provided greater transparency into the decision-making process, allowing HR teams to visualize the factors driving attrition and make informed retention policies.