Workplace dynamics and attrition: A logistic regression approach to understanding employee retention

Abstract

Employee attrition constitutes one of the most consequential and operationally costly challenges confronting contemporary organizations, yet its multifactorial determinants remain imperfectly understood in quantitative terms. This study developed and validated a binary logistic regression model to predict the probability of voluntary resignation among employees, drawing on a secondary dataset of 1,470 employee records obtained from the IBM HR Analytics Employee Attrition dataset via Kaggle. Following point-biserial correlation screening, 19 variables were retained from an initial pool of 35 as significant predictors of attrition. The fitted model demonstrated strong overall performance, achieving 89.0% overall classification accuracy, correctly identifying 97.3% of non-resigning employees, and 46.0% of those who resigned (χ²(41) = 437, p < 0.001; Nagelkerke R² = 0.438). Twelve variables emerged as statistically significant predictors of attrition at the 0.05 level. Overtime work produced the largest positive effect on resignation odds (OR = 7.27, p < 0.001), followed by frequent business travel (OR = 6.52, p < 0.001). Among retention-associated predictors, high job involvement (OR = 0.117 at level 4, p < 0.001), strong environment satisfaction (OR = 0.269 at level 4, p < 0.001), and high job satisfaction (OR = 0.281 at level 4, p < 0.001) were most protective against attrition. Additional significant predictors included distance from home, training frequency, tenure in current role, marital status, job role, relationship satisfaction, and work-life balance. These findings provide an empirically grounded predictive framework for HR practitioners and organizational decision-makers seeking to design targeted, evidence-based retention strategies.