Objective To calculate the nonlinear features of motion in patients with chronic vestibular syndrome (CVS) using the largest Lyapunov exponent (LLE), and to verify the classification model’s validity through machine learning algorithms. Methods A three-dimensional (3D) motion capture system was used to capture the joint motion trajectories of the subjects, which were determined using the LLE. The features of the chaotic trajectories were calculated as the input, and seven classifiers, namely the ID3 decision tree, Adaboost, C45 decision tree, Bayesian classification, Naive Bayes, and support vector machine, were used for classification. Results A total of 17 sets of trajectories from 16 joints were in the chaotic state, and the average energy, enhanced wavelength, and kurtosis of the motion trajectories in the experimental group showed significant differences (P < 0.05). The ID3 decision tree classifier showed optimal performance with 100% prediction accuracy, recall, and F1-score. Conclusions Chaotic features may contain high personality differences in patients with CVS and can improve the accuracy of machine learning algorithms for recognition. These findings provide a reference for early identification and motor rehabilitation of patients with CVS.