Fig. 5: The horizontal bars depicting the results of pairwise comparisons of model performance for each iteration based on feature combinations.

A PPMI database and B FJMUUH-PD database. When comparisons are made, a value greater than 0 indicates that the model for that variable (the clinical variable, the cortical thickness variable, the white matter fiber variable, the combined clinical and cortical thickness variable, and the combined clinical and white matter fiber variable) outperforms the model for the combination of clinical features, cortical thickness, and white matter fiber variables. A value equal to 0 indicates that the performance of the model for that variable matches the performance of the combined model for the clinical features, cortical thickness, and white matter fiber variables. Conversely, if the value is less than 0, it indicates that the performance of the combined model of clinical features, cortical thickness, and white matter fiber variables outperforms the model of this variable. ROSE random oversampling example, SMOTE synthetic minority oversampling technique, SVM support vector machine.