Classification of individual flexibility: before and after a long-term stretching program

Abstract

The evaluation of the stretching effects through mean values of the entire sample can hide different adaptations between groups with distinct characteristics of joint flexibility. This study aimed to employ artificial neural networks to classify subjects according to their flexibility level and to investigate the effects of a 10-week stretching training. The stretching group (SG, n = 9) performed stretching exercises for triceps surae muscles 4–5 times a week over 10 weeks. Maximum dorsiflexion angle (MDA) and peak passive torque (PPT) data at pre-intervention were used in a k-means (k = 3) algorithm to group participants as flexible, intermediate or stiff. A feed-forward artificial neural network (multilayer perceptron) was trained using the pre-intervention dataset with the k-means labeled groups and classified the post-intervention data generated after the stretching protocol. MDA of SG increased significantly (p = 0.015) from 24.72 ± 7.70 to 29.81 ± 6.95° whereas PPT showed no significant differences. Following intervention, two subjects from the SG shifted from the intermediate to the flexible group, whereas one stiff subject changed to the intermediate group. The control group presented a random pattern of group change. This approach can aid future analysis of different adaptations to stretching programs.