*3.3. Training and Testing of SVM 3*

SVM 3 was trained and tested to classify PD1 and PD2 faults. The training stage was conducted by using a total of 8 oil data samples with a PD1 fault condition and a total of 7 data samples with a PD2 fault condition. Therefore, a total of 10 data samples were utilized in the testing of SVM. The results indicated that all the data samples were pigeonholed accurately. The selection of responses and predictors selected by SVM 3 using the ingested data is demonstrated in Figure 11.


**Figure 11.** The dataset variables were selected by the SVM 3 classifier.

The dissolved gases were automatically placed as the predictors and the fault type as the response. In the SVM 3 algorithm, C2H2 and C2H4 were considered to acquire the desired output. The characterization of C2H2 versus C2H4 is illustrated in Figure 12.

**Figure 12.** C2H2 versus C2H4. Blue and orange (•)—raw data, orange and blue (×)—predicted data.

Further, the ML algorithms, were compared in terms of accuracy (%), prediction speed (objects/sec) and training time (sec), as tabulated in Table 6.


**Table 6.** SVM 3 DGA classification outcomes.

In the SVM 3 algorithm, linear SVM, quadratic SVM, and cubic SVM provided the highest degree of accuracy, with cubic SVM yielding 100% accuracy, and the corresponding configuration matrix shown in Figure 13 was imported to the workspace to predict the new testing data samples.

**Figure 13.** Confusion matrix for SVM 3 classifier.
