Machine Learning-Based Detection of Sport-Specific Jumps Using Wearable Inertial Sensor

Wearable inertial measurement units (IMUs) have revolutionized the field of sport performance analysis by enabling continuous and real-time motion monitoring during in-field training settings. This paper proposes an AI-driven solution for sport-specific jump detection using K-AI (K-Sport, Fano, Italy), a wearable IMU sensor integrated into a sports bib designed for athlete monitoring. The proposed solution uses a lightweight Multi-Layer Perceptron (MLP) classifier, optimized for real-time computation, to handle time-series inertial data. The data processing pipeline includes signal filtering, segmentation into fixed-length windows, and identification of movement patterns. The model was trained on 28,856 labeled sport-specific jumps performed by basketball and volleyball players in controlled conditions. The dataset was partitioned into three non-overlapping subsets: 70% for training, 15% for validation, and 15% for testing. The final model achieves an accuracy of 99.85%. Focusing on computational efficiency and biomechanical precision, this approach bridges the divide between laboratory-grade analysis and day-to-day training, real-time feedback for performance optimization, and injury prevention.