Predicting Employee Turnover In Software Companies Using Interpretable Machine Learning Techniques

High employee turnover rate in software and hardware development envi-ronments has a negative impact on organizations or enterprises, causing ad-ditional costs and project delays. This study proposes a predictive ap-proach, based on machine learning, to support employee turnover preven-tion. Especially, it is aimed to obtain a solution that conciliates the predic-tion's accuracy and the models' interpretability, providing useful infor-mation to support the process of decision making at the Human Resources management. To do so, several technical, demographics and behaviour relat-ed attributes were analyzed, and so, a database was built to train and evalu-ate predictive models. Preliminary results indicate the effectiveness of the approach, with satisfactory accuracy rates for the models, as follows: Deci-sion Tree (80%), Logistic Regression (80%), Random Forest (85%), and XGBoost (82%). Regarding the trade-off solution between accuracy and in-terpretability, our analysis reveals that while ensemble methods achieve better predictive performance, intrinsic interpretable models like Decision Tree and Logistic Regression offer competitive and satisfactory accuracy and provide higher structural transparency, having simplicity and good overall comprehensibility. Logistic Regression showed the best trade-off.