Cavitation is the phenomenon which often happened on sentrifugal pump. The impact of cavitation can cause decreasing of pump activities. It might cause the damage if it left unchecked too long. Recently, the popular method used to detect cavitation is vibration signal analysis method. The extraction of statistical parameters using the time domain and frequency domain are the two general basiss in vibration signal analysis. Previous research usually tends to use the statistical parameter feature in time domain or the frequency domain only, so there is still research gap that can be done by combining the time domain and frequency domain. To get more accurate results, further optimization is done using Principal Component Analysis (PCA). PCA can transform vibration signal data linearly and then classify the data, so that it can distinguish normal operating condition and cavitation. The acquisition of vibration signal data was carried out using the Test-rig cavitation scheme by varying the valve cover on the suction side of the pump. Vibration signal data was taken under normal condition with valve fully open condition, level 1 cavitation on valve cover by 25%, level 2 cavitation at 50%, and level 3 cavitation at 75%. Then the data was extracted into 7 features statistical parameter in time domain and 5 statistical parameters of the frequency domain. From 500 data recorded using accelerometer , the data were divided into 400 data training - 100 testing data. Data training was normalized with PCA and produced matrix loading data. Then, the data loading matrix was multiplied by testing data so that the score result was used to classify each test condition. The test results showed, PCA which use combination of features in time domain and frequency domain statistical parameters is not the most optimal method. The most optimal result in detecting cavitation was shown by PCA which used frequency parameters.