Basketball Activity Recognition Using Supervised Machine Learning Implemented on Tizen OS Smartwatch

Abstract

Basketball Activity Recognition (BAR) in sports teams, especially in basketball, to make statistical analysis of player activity data is currently a very important thing. BAR is one part of sports science that recognizes the movement of players in each activity, such as dribbling, passing, etc. Sport science in the sports business is used as one of the factors of coaches and management to determine strategy, starter line-up, check the condition of players after injury, etc. the current technology to recognize player activity only depends on the object detection method of players' through video recordings of players is considered lacking because it only sees the perspective of the coach to reduce players as starter line-up and there is no logical calculation of why players are not installed as starter line-up. One method for recognizing player activity is using a wearable device that has an accelerometer and gyroscope sensor with high accuracy. The values from those sensors will be classified and recognize their activity, i.e., Dribbling, Passing, and Shooting. Smartwatch is one of those wearable devices that meet those criteria. For the activity classification process, the use of the K-NN classification method is the most appropriate because it has a low computational level that is in accordance with the smartwatch specifications. The results of the classification using accelerometer sensor data and gyroscopes with K-NN as an activity recognition method have an accuracy of 81.62%, and player activity recognition applications using accelerometer and gyroscope sensors can also record the results of player movements for further analysis by management and coaches. This is the advantage of this BAR application compared to the recognition of player activity using object detection on video recordings.