Classification of success/failure in weightlifting snatch using a machine learning model based on barbell kinematics

Abstract

The purpose of this study was to classify and predict success and failure in weightlifting snatch lifts using the barbell kinematics through machine learning, and to identify the key factors that have the greatest impact on success and failure, contributing to performance improvement. Videos from the Korea National Weightlifting Championships from 2022 to 2024 were collected, involving weightlifters from weight classes (male: 55∼+109 kg; female: 45∼+87 kg). The data analyzed included 579 trials from male weightlifters and 472 trials from female weightlifters. Fifteen barbell kinematic variables were analyzed and classified into success and failure categories to extract classification accuracy of machine learning models through multicollinearity. A total of eight machine learning models were used in the analysis. And the most discriminative barbell kinematic variables were identified through feature importance analysis. The machine learning models’ classification accuracy showed the Random Forest for males performed best at 77.72%, while the Support Vector Machine for females achieved 75.38%. For the male & female, the Random Forest achieved 80.77% accuracy. Feature importance analysis revealed dy3 (maximum height to the catch-vertical displacement) as the most significant for both genders, followed by pvV (peak vertical velocity) and dy2 (start to the catch-vertical displacement). In classifying success and failure in the weightlifting snatch lifts, the RF model showed the highest classification accuracy for the male and male & female groups, while the Support Vector Machine performed best for the female group. The dy3 variable had the greatest impact on classifying success and failure.