Abstract: Maintaining a data storage device having a plurality of physical storage units includes constructing a model that predicts failure of the physical storage units in the data storage device, where the model is trained using empirical data containing operational parameters of physical storage units, determining failure predictions of the physical storage units in the data storage device using the model, moving applications in the data storage device based on the failure predictions of the physical storage units in the data storage device, and maintaining the physical storage units in the data storage device according to the failure predictions of the physical storage units in the data storage device. The model may be constructed using a Random Forest Classifier model from the Scikit-Learn software package. Using empirical data containing operational parameters of physical storage units may include accessing publicly available empirical data.