Induction machines are commonly used in the industrial applications. They require continuous monitoring to prevent the premature failure, of which bearing faults account for most of the causes. The bearings in the rotary equipment tend to suffer fatigue failures during certain conditions such as load variation or electrical discharge. The fault detection based on the vibration analysis could be found in lots of previous literatures, and the frequency spectrum was usually adopted to identify the fault characteristic frequency. However, previous method could only detect the occurrence of bearing faults rather than making the prediction. Hence in this study, the time and frequency features were extracted to construct regression models to predict the development of the bearing faults.