Comparison of random forest and support vector machine for prediction of cognitive impairment in Parkinson's disease

Abstract

Cognitive impairments are typical in PD and are indicated by mild cognitive impairment (PD-MCI) in the early stages or dementia (PDD) in higher stages. The Montreal Cognitive Assessment (MoCA) is an instrument commonly used for ascertaining cognitive impairments in PD. This research uses the clinical, neuroimaging, and CSF data as a predictor variable and the MoCA score as the target variable representing cognitive impairments. Machine learning approaches through support vector machine (SVM) and random forest (RF) methods were applied for modeling. The mean absolute error (MAE) and the root mean square error (RMSE) are used to compare the predicted performance values of the method's application. The experimental results showed that both SVM and RF performed well in predicting cognitive impairments in PD patients, indicated by the relatively small MAE value at 0.076 and RMSE at 0.542. This research also discovers that SVM is better than RF in predicting cognitive impairments. Meanwhile, RF presents an apparent and explicable outcome, which is beneficial for determining important variables that correspond to cognitive impairments. The five measurements with the highest mean decrease accuracy (%IncMSE) are age of onset, phosphorylated tau, α-synuclein (aSyn), mean putamen, and total tau.