Purpose The aim of this study is to develop a method for predicting voluntary employee turnover using grey machine learning and to demonstrate its effectiveness as a decision-support tool in HR management. Design/methodology/approach This study presents an innovative approach that integrates machine learning with grey system theory to predict voluntary employee turnover. The developed method follows a six-step procedure, encompassing variable selection, data preparation, and model training. A grey cluster-based decision model is applied as an activation function within a novel machine learning architecture, enabling the incorporation of imprecise, vague, and error-prone data. A quantitative study was conducted using survey data from 1,030 employees, collected through a CAWI method. The dataset was analyzed using machine learning techniques, and model validation was performed through cross-validation. This methodology enhances HR decision-making by addressing uncertainty in turnover prediction. Findings The study demonstrates that integrating grey system theory with machine learning enhances the prediction of voluntary employee turnover by effectively handling imprecise and uncertain data. The developed grey cluster-based decision model improves the representation of social science complexities, capturing key turnover factors such as salary, job satisfaction, and organizational culture. The model’s performance was validated through a quantitative study, achieving an average accuracy of 80.94%. These findings confirm its applicability in HR decision-making, suggesting that grey machine learning provides a robust framework for predicting employee departures and enabling organizations to implement proactive retention strategies while minimizing turnover-related costs. Originality/value The originality of this study lies in the unique integration of grey system theory with machine learning architecture through the application of a grey cluster-based decision model as an activation function. This approach has not been widely applied to predicting voluntary employee turnover, and its novelty stems from the ability to operate directly on imprecise and incomplete data without the need for full “whitening”. The developed model captures complex, multidimensional, and often subjective factors influencing employee decisions, making it more resilient to informational noise than traditional algorithms. An additional value is the method’s potential transferability to other areas of social science research where uncertainty and incomplete data pose significant challenges, opening opportunities for further development in both theoretical and practical analytical contexts.
Marcin Nowak (Wed,) studied this question.