Evaluation of sparse inertial sensor configurations for tennis stroke classification: balancing performance and wearability

Accurate classification of stroke movements is important for performance evaluation in rotational racket sports such as tennis. This study compared the classification performance of six lower-body sensor configurations to assess the feasibility of sparse inertial measurement units (IMUs) setups for tennis stroke classification. Twelve players performed five stroke types, while five IMUs recorded tri-axial acceleration from the pelvis, thighs, and shanks. A one-dimensional convolutional neural network (1D-CNN) with data augmentation was evaluated using subject-level 4-fold cross-validation. The configuration using two sensors on the pelvis and right shank achieved up to 88.17% ± 3.69% accuracy, close to that of the full five-sensor configuration. These findings suggest that, among the six predefined configurations evaluated, sparse IMU configurations with augmentation may provide a practical balance between classification performance and wearability for tennis stroke classification.