<![CDATA[Over the past three years, Perseroan Luar Negeri Ltd.’s five Singapore-based subsidiaries have experienced a consistent increase in attrition rates above annual average benchmarks, raising concerns about service continuity, increased rehiring and onboarding costs, and weakened customer relationships. To address this, the Group implemented machine-learning models designed to potential leavers six months in advance. This research evaluates the first full year of the model's deployment and proposes integrating predictive insights into HR decision-making. The evaluation includes both offline assessments (precision-recall screening of six algorithms during production) and an online evaluation using Wilson-scored recall/precision, Popular-Stability-Index for covariate drift, Linear Four Rate for concept drift, and two business KPIs: voluntary turnover delta and retention yields. Findings shows that only recall-focused models met business targets. Covariate drift, likely triggered by performance-rating freezes and mandated training, caused significant recall deterioration, whereas concept-drift tests remained negative, validating algorithm logic. To address these issues, the study proposes short-term solutions through model retraining. Mid-term actions include conducting exit-interview analyses, redefining attrition baselines, developing quality-control dashboards, establishing comparative benchmarks, and adjusting voluntary turnover calculations to highlight hidden successes during market volatility using industry attrition projections and sector-engagement indices. For long-term sustainability, the study recommends comprehensive training and documentation programs to establish robust talent-risk governance.]]>
Copyrights © 2025
