This study introduces a fast and accurate nonperiodic short-term predictor, ANFISNCH, as a specified web services for forecasting the flow of data packets between server and clients. Even though ANFIS is a fast fuzzy inference or predictor, the phenomenon of volatility clustering always generates the extreme outliers embedded in the training data set because of the effect of nonlinear conditional heteroscedasticity, and ANFIS in fact cannot overcome this problem resulted in a trained model that is not the optimal one. ANFISNCH model employing segmented adaptive support vector regression (SASVR) learning algorithm to adjust between ANFIS output and nonlinear conditional heteroscedasticity can best fit the model and greatly reduces the occurrence of extreme outliers in the predicted outputs from ANFISCH.