Fig. 5: Machine-learning performance and normalized feature-importance profiles for MMSE and MoCA prediction. | npj Parkinson's Disease

Fig. 5: Machine-learning performance and normalized feature-importance profiles for MMSE and MoCA prediction.

From: Classifying the risk of cognitive impairment in Parkinson’s disease using serum bile acid profiles and machine learning

Fig. 5: Machine-learning performance and normalized feature-importance profiles for MMSE and MoCA prediction.The alternative text for this image may have been generated using AI.

A ROC curves for classifying MMSE ≥ 27 versus <27 based on our 7 BA panel (DCA, LCA, GLCA, TCA, CA, TLCA and GCA), using five classifiers: RBF SVM, linear SVM, RF, XGBoost and MLP. B Absolute-residual boxplots showing the distribution of the prediction error (observed minus predicted probability) for each MMSE model. C Reverse-cumulative residual curves for MMSE models, depicting the proportion of predictions within increasing error thresholds. D Stacked bar chart of normalized importance scores (0–100%) for each BA across the five MMSE classifiers, based on RF mean-decrease-Gini, XGBoost gain, and absolute weight magnitudes for the SVMs and MLP. E ROC curves for classifying MoCA ≥ 26 versus <26 with the same 7 BA panel and classifiers. F Absolute-residual boxplots for MoCA models. G Reverse-cumulative residual curves for MoCA predictions. H Stacked bar chart of normalized importance scores for each bile acid across the five MoCA classifiers.

Back to article page