Employee Turnover Modeling Using Feature Importance Through Logistic Regression in the Security and Operational Service Sector

Abstract

Employee turnover has become a major challenge for many companies, including in Indonesia, which recorded an average rate of 41% in 2023, disrupting the effectiveness and smoothness of company operations. This study aims to predict employee turnover using the Logistic Regression algorithm. The process includes data preprocessing, data splitting for training and testing, modeling, and evaluation. The model achieved an accuracy of 90.3% and an F1-score of 0.81 using an 85:15 train-test split, indicating strong predictive performance. The most influential factors identified through feature importance analysis are year of joining, length of service, position, allowance, and replacement status. Visualization results show that employees with shorter tenure and more recent joining years are more likely to leave, suggesting the need to improve orientation, training, and work adaptation support. Logistic Regression proves effective in assisting the Human Resources division in designing data-driven retention strategies and supporting decision-making in workforce management.