When [k.sub.p] = [k.sub.o], the prediction error of KNN-S2S guarantees the best case of SKNN within +0.2%.
With regard to a suitable [k.sub.p] value, if the [k.sub.p] value is sufficiently larger than the [k.sub.o] value, reliable prediction accuracy is at least guaranteed by approximating the best case of SKNN within an acceptable error margin, which is insensitive to the prediction accuracy.
In addition, the prediction error of KNN-D is exactly the same as that of SKNN with [beta] [right arrow] 0.5 x [S.sub.n].
Based on the results of a parameter analysis, the optimal values of the model parameters for SKNN and KNN-S2S were determined to be [[d.sub.o] = 5, [k.sub.o] = 41, [k.sup.o.sub.p] = 400] for result analysis.
The two best performers were KNN-S2S and SKNN, which are followed by the two second-best performers, KNN-UD and KNN-U.
The conformity of predictions between SKNN using a whole dataset and a KNN model combined with a segmentation method is very crucial.
However, computational loads equal to those of SKNN to search through all of the historical data are inevitable during nonprediction time (i.e., time interval (t)).