Bearings are widely used in rotatory machines which are running continuously for many hours. Knowing the remaining life time of the bearings will help to schedule the preventive maintenance program and to buy the necessary spare parts in advance. Researchers generally use regression techniques to assess the remaining life time of the bearings. This paper proposes a machine learning based classification approach to assess the remaining life time of bearings. The vibration signals were acquired on continuous basis from bearings operating rated speed and load conditions. The experiment was carried out on a new bearing until the bearing is damaged (not suitable for use). The time duration from start of the experiment to the end of the experiment is divided into five stages. From the sample signals of these stages, descriptive statistical features like mean, median, skewness, etc., were extracted. Out of 12 such features, the best performing five features were selected using decision tree. Then the selected features were used for training the classifiers to build a predictive model. The trained decision tree algorithm (classifier) was validated using distinct data. The classification performance is 95.48%. © Research India Publications.