Using our classification method, we decided to test our accuracy on 7 different subjects. Four fingers were used during these tests (index, middle, ring, and little finger). For the accuracy test, we asked the subject to perform each of the movements (rest, pinch, point and power grasp) repeatedly 20 times each and finally one extra repetition of rest. In total, a given subject would perform 81 different movements. For all subjects, we repeated the accuracy test three times on separate days in order to assess the accuracy of our code.
From these tests we got the following data:
We then used a confusion matrix to graph the accuracy of our data. The confusion matrix can be found below.
From the confusion matrix below it can be seen that the overall classification accuracy is 89.5 percent which can be seen at the bottom right corner in purple. The green diagonal corresponds to when an output class is being classified correctly as it’s target class. The red areas show where a class can be misclassified. As can be seen in above, the majority of the data is being correctly classified. The gray areas show the total accuracy per row and column. From this, we can learn that rest is never misclassified as power grasp and that rest and pinch can be misclassified for some subjects.