Fig. 2: SHAP summary plot shows feature importance for distinguishing patients with cognitive impairment from CU controls.

Positive SHAP values indicate a higher likelihood of classification as a patient. Red represents higher values for each feature, while blue represents lower feature values. LII was the strongest predictor, with low language informativeness values increasing the probability that the individual was a patient with cognitive impairment. LII was followed by pause duration (high values make cognitive impairment more likely), number of sentences (high values make cognitive impairment more likely), and word length (low values make cognitive impairment more likely).