Exploring Sex Differences in Predictors of Jump Height: Linear Lasso Machine Learning Feature Selection
Jason Jagroop
Co-Presenters: Individual Presentation
College: The College of Health Professions and Human Services
Major: Excercise Science (M.S.)
Faculty Research Mentor: Pragya Sharma Ghimire
Abstract:
Countermovement vertical jump (CMVJ) is recognized as a strong indicator of lower-body strength and performance. With the increasing availability of commercial force plates, key metrics such as the reactive strength index (RSI) are now more accessible than ever. Purpose Sparta Force Plates offers comprehensive and unique performance metrics for detailed assessments of athletes’ strength, explosiveness, and balance. This study assesses and distinguishes the significance of Sparta’s jump metrics in forecasting CMVJ using the Linear Lasso (LL) machine learning (ML) algorithm. Methods Data were collected from 233 collegiate athletes, consisting of 56.7% males and 43.3% females, across 11 distinct sports. Following a warm-up trial, participants performed 3 CMVJs on the Sparta Force plate, with the best trial selected for analysis. Before the analysis, the dataset was stratified by sex. Twenty-eight different jump measures, height, weight, body mass index, and age were entered as predictors of CMVJ height in the LL analysis. The LL algorithm was trained and tested on 70% and 30% of the data, respectively. Five-fold crossvalidation was performed on the training set and fine-tuned using a grid of alpha values from 0.01 to 2.0. The model with the best predictive value was selected. Results The predictors of CMVJ retained for males included maximum velocity (0.756), average maximum force (-0.004), maximum power (0.024), countermovement depth (0.007), jump height flight time (0.036), COPy unweighting displacement (0.006), and mRSI (0.014). The model achieved a test set R2 of 0.994. For females, the LL model retained maximum velocity (0.453), average maximum force (0.001), jump height flight time (0.039), average relative concentric force (0.003), and mRSI (0.012). The model achieved a test set R2 of 0.997. Conclusions: The most important predictor in males and females was maximum velocity. Although some predictors were shared, there were some distinct differences in the feature sets predictive of CMVJ among males and females. Significance/Novelty: Utlizing this LL method could help to identify in selecting important predictors of CMVJ and injury risk, allowing sport performance coaches to develop highly specific training programs.Key Words: Countermovement Jump, Injury, and Machine Learning Algorithm