8.2.2. Evaluation Metrics

To validate the skill of the proposed classification model, five-fold cross-validation was applied, which is a typical resampling technique that shuffles the dataset randomly and splits it into five equal-sized groups. From those, four groups were used for training the model, and one group was kept for testing the model. Five iterations were performed with the grouped data to cross-validate the model by using each group as a testing dataset while employing the others for training. The average of the evaluation metrics from the five iterations was taken as the final evaluation score of the model. To evaluate the motion mode classification performance, several metrics were used in this study, including accuracy, precision, sensitivity, specificity, and F-measure. The accuracy for a motion class is the ratio of correctly labelled motion modes for that class to the total number of labelled motion modes for that class. The precision for a motion class can be defined as the ratio of correctly positive-labelled motion modes for that class to the total number of positive-labelled motion modes for that class. The sensitivity (which is also known as recall) for a motion class is the ratio of correctly positive-labelled motion modes for that class to the total number of motion modes that actually belong to that class. The specificity for a motion class is the ratio of correctly negative-labelled motion modes for that class to the total number of motion modes that do not belong to that class. F-measure is the harmonic average of precision and sensitivity. These metrics are defined as follows:

$$Accuracy = \frac{TP + TN}{TP + FP + FN + TN} \tag{21}$$

$$Precision = \frac{TP}{TP + FP} \tag{22}$$

$$Sensitivity = \frac{TP}{TP + FN} \tag{23}$$

$$Specificity = \frac{TN}{TN + FP} \tag{24}$$

$$F-measure = 2 \ast (\frac{Precision \ast Sensitivity}{Precision + Sensitivity}) \tag{25}$$
