4.1.6. Support Vector Machine

The support vector machine (SVM) algorithm was optimized by tuning the kernel and filter type parameters to improve its performance. The results obtained are presented in Table 7. We found that a combination of the poly kernel and the normalization of the training data resulted in the least error rates across both RAE and RRSE. In this case, both the polynomial and normalized polynomial kernel combined with normalization of the training data achieved the same performance. However, we note that it will be of greater value computation wise to avoid the normalization expenses of the poly kernel, thus implying that using the simple poly kernel should suffice for the present case. Similar to the GR algorithm, the RBF kernel yielded the largest error rates with the same plausible reasons as stated for the GR algorithm applying as well to the SVM algorithm. Summarily, an average of 38.15% reduction in the error rate was achieved by using the poly kernel over the RBF, thus reemphasizing the importance of hyperparameter tuning in the use of ML algorithms.


**Table 7.** Performance of different SVM parameter settings.
