An improved baseline acquisition approach for real-time performance monitoring based on support vector machines

To acquire the baselines of key performance indicators (KPI) which are critical for the real-time performance monitoring (RTPM), an improved time series prediction approach is proposed based on support vector machines (SVM). Considering the characteristics of the KPI time series, wavelet multi-resolution is carried before modeling by SVM, and the result is the sum of prediction values of each branch. Experimental results show that the prediction is of high precision.