Principal Components of Superhigh-Dimensional Statistical Features and Support Vector Machine for Improving Identification Accuracies of Different Gear Crack Levels under Different Working Conditions

Table 3

Comparisons of the prediction accuracies obtained by using the proposed method and the four methods reported in [16] (unit: %). (Note: methods 1 to 4 are used to distinguish three gear health conditions, while the proposed method is used to classify five gear health conditions.)

Prediction accuracy	Experiment 1	Experiment 2	Experiment 3
-nearest neighbor without statistical feature selection [16]	89.58	87.42	87.67
-nearest neighbor with random statistical feature selection [16]	80.17	77.29	78.08
-nearest neighbor with Euclidean distance evaluation [16]	96.53	99.15	92.10
The weighted -nearest neighbor method [16]	96.68	99.61	92.62
The proposed method	100	100	100