Centrifugal pump is one type of pumps that widely used, especialy in industry. It’s mechanism which cretates pressure changes usually caused cavitation. Generally, centrifugal pump fault caused by cavitation makes high noise and vibration level. Cavitation phenomenon that is not properly maintain results fatal breakdown and high economic losses. Therefore, research is needed to find and develop the method that can detect early cavitation phenomena in centrifugal pumps, and can show cavitation at several levels. An early cavitation detection can be done by vibration signal analysis using Support Vector Machine (SVM) based method. Data with sevel variety (normal, level 1, 2, and 3 cavitation) was extracted into ten statistical features. Then, it was also selected using Relief Feature Selection. In this study, two types of SVM method were used to calssify the data named binary and multi class SVM. Each multi class classification of result is optimized by Bayesian Optimization (BO) algorithm and Grid Search Method (GSM). The whole processes was carried out using Matlab R2107a. The results showed that each statistical feature contained spesific information on vibration data. Root Mean Square (RMS), Standard Deviation (SD), variance, entropy, and Standard Error (SE) are several features that showed the best plot. Feature selection process revealed that variance, RMS, and SD were the best feature to use for SVM classification. Binary SVM method showed the best plot on early cavitation with accuracy 99%. BO algorithm with multi class SVM was the best combination to classify all varieties with overall accuracy 100%.