Light sport exercise detection based on smartwatch and smartphone using k-Nearest Neighbor and Dynamic Time Warping algorithm

Implementation of human activity recognition in daily life has been widely applied, one of which is for the health sector. This research proposes a light sport exercise activity detection system that can be easily done by person. Among these activities are walking, push up, sit up, and squat jump. The system is able to recognize the movements made by the user and calculate the number movements. This research involved 25 respondents, 15 samples were used for training data, and 10 samples for the test data. The system uses the accelerometer sensor on a smartphone and smartwatch that is placed on the left hand of the user. Time series data from accelerometer value will be process by sliding window method with k-Nearest Neighbor algorithm and Dynamic Time Warping as a main classification algorithm. In the feature extraction process, walking activity is not entered into the system since it has random pattern. Thus, only push up, sit up, and squat jump activities are classified into the system. On the value of k = 3, the accuracy of push up motion is 76.67%, then 80% for sit up, and 96.67% for squat jump activity. Based on this result, we can conclude that k-Nearest Neighbor and Dynamic Time Warping can classify the light sport exercise activity.