AN AI BASED SOLUTION TO ASSESS EXECUTIVE EMPLOYEE TURNOVER RISK IN THE APPAREL INDUSTRY

Abstract

ABSTRACT Employee turnover, particularly among executive and above cadres, presents a significant challenge in the Sri Lankan Apparel industry, impacting organizational stability and productivity. This thesis aims to address this issue by predicting the likelihood of employees leaving their current positions within the next year and estimating the probable time frame for such turnovers. Initially, a descriptive analysis of executive and above employee behavior within a major apparel manufacturing company revealed a concerning trend of increasing turnover among long-term and skilled employees over the past five years. Factors such as prolonged tenure within a single grade correlated positively with turnover rates, while employees with diverse job roles exhibited lower turnover tendencies. These findings underscored the urgent need for proactive measures to mitigate turnover risks. Transitioning to predictive modeling, we formulated turnover prediction as both a binary classification problem to ascertain turnover possibility and a multi-classification problem to predict turnover horizons. Leveraging supervised machine learning techniques and publicly available employee data from LinkedIn, we trained models to forecast turnover events. The XGB Classifier emerged as the most effective algorithm, achieving accuracies of 81% for turnover possibility prediction and 75% for turnover horizon estimation. Key features influencing turnover likelihood included the frequency of internal promotions, tenure, job durations, and educational qualifications. These insights emphasize the importance of continuous monitoring of such variables to preempt turnover events effectively. Furthermore, we developed a user-friendly interface to facilitate easy access to turnover risk scores and timelines based on employee LinkedIn profiles. In considering future research directions, we propose integrating internal data sources to enhance model accuracy and exploring additional variables such as salary and employee feedback. Moreover, the analysis can be extended to encompass internal turnovers and industryspecific turnover patterns, offering tailored insights for diverse organizational contexts. Additionally, incorporating social network analysis and exploring turnover prediction from a company-wide perspective present promising avenues for further investigation. Overall, this study provides valuable insights and tools to proactively manage employee turnover in the apparel industry and beyond.