Prediksi Turnover Karyawan Berbasis Data Perilaku Kerja pada Karyawan Freelance

Abstract

This study aims to develop a predictive model of freelance employee turnover by integrating digital work behavior data and psychological perceptions, as the high turnover rate in the freelance sector has a significant impact on the sustainability of digital platforms. A total of 150 freelance respondents from various fields in South Sumatra participated in this research. Data were collected through digital activity logs and perception surveys, including measures of job satisfaction, work stress, and affective commitment. Quantitative analysis was conducted using logistic regression and the Decision Tree (C4.5) algorithm to construct a predictive model of turnover intention. The findings indicate that work stress has a significant positive effect on turnover intention, while job satisfaction, affective commitment, and digital communication frequency have significant negative effects. The Decision Tree model demonstrated strong performance with an accuracy of 82% and an AUC of 0.85, identifying work stress and communication frequency as dominant factors in turnover prediction. These results support the Affective Events Theory and Perceived Organizational Support, showing that the combination of psychological and data-analytic approaches can serve as a foundation for developing early-warning systems for turnover in digital workforces. This study contributes to the advancement of psychology-based HR analytics within the growing gig economy in the digital era.