Machine Learning-Based Prediction of Employee Turnover in Iraq’s Private and Engineering Sectors

Abstract

It forecasts staff turnover in the private and engineering sectors of Iraq using machine learning methods. The analysis of labor market conditions in Iraq, which exists in conditions of prolonged economic instability, labor mobility, which reveals the problems of staff retention in engineering, manufacturing and technical service companies, justifies the need for proactive decision support tools. Applied research was conducted on the basis of the public dataset Human Resource Analytics. Unlike most studies using the dataset in its original form, a methodology for expanding and transforming factors is proposed. In particular, the subjective indicator of employee satisfaction, rarely observed in real organizational conditions, has been eliminated and a set of derived and integral indicators formed reflecting productivity dynamics, loyalty, workload intensity, seniority effectiveness, professional growth, and stress-related factors. Special attention is paid to the development of the SSI integral indicator (Satisfaction Score Index), which acts as a substitute for the employee satisfaction indicator and allows maintaining the predictive ability of the model. Feature scaling, SMOTE class balancing, cross-validation, comprehensive assessment using ROC-AUC, PR-AUC, and F1-score metrics, as well as isotonic regression for probability calibration to provide reliable and interpretable risk estimates were performed. The experimental results showed that ensemble algorithms (Random Forest, Gradient Boosting) demonstrate the best predictive effectiveness, and the formed new features are identified as the most influential associated with staff turnover. While the study is motivated by the peculiarities of the Iraqi labor market, the methodology proposed can be generalized to private and engineering organizations in other regions under the same conditions of relocation, burnout, and high turnover. The approach proposed can be taken as a basis for the implementation of intelligent systems for predicting the risk of dismissal in HR analytical platforms.