A machine learning-based model for athletic ability age assessment by gait characteristics

Background: Age is one of the key factors that significantly affects body movement balance.Research question: Can age prediction be performed by training machine learning models using machine vision-based gait data?Methods: A total of 96 subjects of varying ages were recruited for this study, and gait data were collected from each subject as they walked continuously on a treadmill for one minute. Features such as period and amplitude were extracted. These features were compared with different feature selection methods (recursive feature elimination, maximum information coefficient, and mutual information), and the optimal feature matrices were input into four different classification models, including random forest, logistic regression, support vector machine, and decision tree for 4-level classification.Results: The performance of four classification models improved after feature selection, and the Random Forest model with feature selection of mutual information excelled in effectively recognizing gaits of different age groups, with an accuracy of 84.21%, precision of 87.5%, recall of 79.76%, and F1 score of 80.58% using only three features.Significance: The results show that machine vision combined with machine learning can correctly classify different age groups using a small number of features, which is of screening significance for people with premature gait and potential gait risks.