According to the three different similarity measurements, the proposed clustering-based forecasting scheme can
be viewed as three forecasting models. They are Mean-F model using Mean method, Median-F model employing
Median method and MinMax-F model applying MinMax method. The prediction results of the proposed three sales
forecasting schemes are compared to those of SVR model without data clustering (called single SVR model).
For building forecasting models, five predictors employed in [7,8] including previous month’s sales amount (T-
1), previous two months’ sales amount (T-2), previous three months’ sales amount (T-3), 3-month moving average
(MA3) and sales amount at the same time last year (T-12) are used in this study. In this study, all of the four
forecasting schemes are used for one-step-ahead forecasting of monthly sales data. The prediction performance is
evaluated using the root mean square error (RMSE), mean absolute deviation (MAD), mean absolute percentage
error (MAPE), root mean square percentage error (RMSPE) and normalized mean square error (NMSE).