Objective: To evaluate the performance of a Random Forest (RF) classifier on Transcranial Magnetic Stimulation (TMS) measures in patients with Mild Cognitive Impairment (MCI). Methods: We applied a RF classifier on TMS measures obtained from a multicenter cohort of patients with MCI, including MCI-Alzheimer's Disease (MCI-AD), MCI-frontotemporal dementia (MCI-FTD), MCI-dementia with Lewy bodies (MCI-DLB), and healthy controls (HC). All patients underwent TMS assessment at recruitment (index test), with application of reference clinical criteria, to predict different neurodegenerative disorders. The primary outcome measures were the classification accuracy, precision, recall and F1-score of TMS in differentiating each disorder. Results: 160 participants were included, namely 64 patients diagnosed as MCI-AD, 28 as MCI-FTD, 14 as MCI-DLB, and 47 as healthy controls (HC). A series of 3 binary classifiers was employed, and the prediction model exhibited high classification accuracy (ranging from 0.72 to 0.86), high precision (0.72–0.90), high recall (0.75–0.98), and high F1-scores (0.78–0.92), in differentiating each neurodegenerative disorder. By computing a new classifier, trained and validated on the current cohort of MCI patients, classification indices showed even higher accuracy (ranging from 0.83 to 0.93), precision (0.87–0.89), recall (0.83–1.00), and F1-scores (0.85–0.94). Conclusions: TMS may be considered a useful additional screening tool to be used in clinical practice in the prodromal stages of neurodegenerative dementias.
