A machine learning framework for evaluating basketball player skill development using video analysis

A lot of time and effort is used in the conventional approaches to measuring the performance of basketball players. The paper will deal with the necessity of an improved and precise system to assess the skill development of basketball players. The proposed approach uses three-phase process. First, player tracking is done by applying background subtraction and IMM to deal with occlusion problems. The second phase is action recognition by using 3D Convolutional Neural Network (CNN), and finally in the third phase of the proposed method, the player skill is assessed by using ensemble model. This combined model is a new way of automating performance assessment. The model was tested using a set of real NBA match videos and the accuracy of the model was found to be 88.34% for action recognition and 93.19% in evaluating player skills. This goes to show how the proposed approach works well in identifying the various actions of the players and assessing the level of improvement. The proposed framework could be useful for the coaches and analysts to have a better way of evaluating the performance of the players and the training that is to be done. Finally, this framework offers a powerful and effective platform of objective analysis of player performance and training optimization.