Fatigue Classification by ML_2paper

To enhance the interpretability and predictive performance of the classification models, this study first conducted paired sample t-tests on all eye-tracking features to identify those with statistically significant differences before and after visual fatigue. A total of eight features were found to be significant (p < 0.05), including four linear parameters (FC_10, FC_25, OL_10, OL_25), two spatial parameters related to gaze dispersion (Radius_50 and Dist_CM), and two nonlinear complexity indicators (SamEn_x and SamEn_y). These selected features encompass multiple dimensions, including fixation density (fixation count), fixation duration (observation length), spatial dispersion of gaze (fixation dispersion), and signal complexity of gaze trajectories (Sample Entropy), providing both behavioral-level and signal-structure-level insights into changes associated with visual fatigue. In the subsequent AI classification phase, three types of models were constructed: one based on linear features (FC_10, FC_25, OL_10, OL_25, Radius_50, Dist_CM), one based on nonlinear features (SamEn_x, SamEn_y), and a hybrid model combining both linear and nonlinear features. These feature sets were input into three supervised machine learning algorithms—Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)—for comparative analysis. Model performance was comprehensively evaluated using metrics such as AUC, Accuracy, Precision, Recall, and F1-score, in order to assess the contribution and discriminative power of different feature types in identifying visual fatigue.