Recently, a renewed interest in adopting data analytics to help solving HR problems and to make more informed and effective choices appeared in the literature. One of the greatest challenges for organizations is employee turnover because of its adverse impact in many areas, such as productivity, performance and reputation. In case of attrition, one of the problems is that data from HRIS are complex, full of sensitive information (GDPR) and of useless data. Once data are clean, they can be analyzed by using statistical approaches such as machine learning. This study is about predicting employee attrition using machine learning models on a real dataset of a large Italian financial company, and, in particular, we focus on choosing the best. This contrasts with much extant research which is based on artificial datasets. To address this issue, machine learning tools have been developed for investigating and predicting employee attrition, as well as methods for evaluating their predictive power. Evidence on what are the most important predictors that lead to attrition and in what areas it is more likely to happen enable HR managers to implement targeted retention policies and practices. The contribution of this paper is to explore and compare the performance of several common models which are found in the literature on real data. Then, we focus on the results of the best performing model and identify some groups of employees who have a high risk of attrition on which the company could intervene to reduce voluntary resignation.