This study classifies the actions of football players using sensing data acquired from wearable sensors attached to players and the ball. More than 800 sensing data with the labels of five types of player actions were created as a dataset. The neural networks were trained using 19 input items created by considering time-series variations in player and ball locations. The trained neural network model demonstrated a classification accuracy of 84.0%. The model successfully obtained sufficient accuracy for all types of actions.These results demonstrate that the sensing data and created input items can be effectively utilized for classifying the actions of football player.
