Advanced Kinematic Analysis of Cricket Bowlers Using Wearable Sensors and Machine Learning

Fast bowlers in cricket face the constant challenge of intense physical actions, which makes them susceptible to injuries. Kinematic analysis is crucial for evaluating a bowler’s skills, identifying strengths and weaknesses, reducing the risk of injury, and enhancing performance. This study presents the kinematic analysis of the cricket bowler using indigenous miniature and cost-effective wearable sensor devices. The proposed wearable device has dimensions of $20\times 23\times 6$ mm and weighs only 6 g. Given that bowlers are often 60–70 m away from camera setups on the pitch, these wearable devices provide a practical solution for recording motion data. The proposed wearable device utilizes the Bosch BNO055 system in package (SiP) as the embedded sensor and is mounted on the player’s arm and back to analyze upper body kinematics. A novel method for ball count and ball release detection is presented using data from both arm-mounted and back-mounted devices. The back-mounted sensor analyzes back flexion during ball delivery. Orientation information extracted using quaternion algebra, and statistical features extracted from the sensor data have been used to evaluate supervised machine learning algorithms for classifying bowling phases. Machine learning models, including decision tree, k-nearest neighbor (KNN), and neural network, were trained to classify bowling phases. The ensemble bagged tree model achieved the highest accuracy of 98.1%, with precision, recall, and ${F}1$ scores of 92.5%, 90.4%, and 91.4%, respectively. The KNN and bilayered neural network models demonstrated accuracies of 96.9% and 96.7%, while the wide and trilayered neural network models achieved accuracies of 96.5%.