Centrifugal pumps are very essential components for fluid transfer in the industry and in self-priming centrifugal pumps, the bearing is the key element. The performance of the pumps highly depends on the bearing conditions and high-performance leads to smooth operation and lesser downtime. But with time when defects introduce in the bearing, the detection of the type of defect is the major challenge. Conventional ways of detecting defects lead to high maintenance costs and time. For the detection of defects in the bearing, vibration signal analysis is widely used in the last decade and this research also focuses on the Signal processing techniques for such as Discrete Wavelet Transform (DWT) and Modulated Wavelet Packet Transform (MODWPT). Along with this, soft computing techniques, including Ensemble Bagging Tree Algorithm (ETA), Decision Tree Algorithm (DTA) and Support Vector Machine Algorithm (SVMA) are compared to make the classification more intelligent and to make this approach adaptive for fault detection. Due to the advantages of all these techniques in nonlinear problems, the results are impressive in comparison. A data set is first denoised using signal processing in which four different wavelets are selected and using the best-performing wavelet, thirteen different Statistical Time-Domain Features are extracted for training the models. Results show that the best performance is achieved using MODWPT and SVMA