Voluntary employee turnover is an essential threat to companies due to the loss of institutional expertise and costs associated with recruitment. In this paper, we continue to tackle the retention problem by developing a machine learning (ML) based solution and providing a prototype tool to predict potential turnover. Both unsupervised and supervised ML methods are applied to identify the appropriate technique. K-Means clustering algorithm with PCA did not show significant results, whereas CART decision tree algorithm reached 89 % accuracy on the test set and 84 % accuracy on the validation set. Having satisfactory classification results are only part of a successful solution for modelling human resource (HR) dynamics. An important yet often underestimated aspect is representation according to end-user needs, which rarely are ML experts. To mitigate the risks of misinterpretation of classification results and take full advantage of decision support, we emphasize engineering the output of classification results for HR employees. We validated the classification system on data sets containing records of more than 2000 employees working in technology companies in Latvia from the year 2014 up to date.