FIG. 1.

OOB error rates were recorded after each dimension reduction. DR below 100 features was only continued in run 3, for exemplary purposes and comparison with Yahata and colleagues (2016). In all three runs, error rates decreased as the number of features was reduced. This trend was reversed when the models incorporated fewer than 200 features, suggesting that important diagnostic information is lost when models become oversimplified. The model that achieved the peak accuracy of 95.6% used 400 features. OOB, out-of-bag; DR, dimensional reduction. Color images available online at www.liebertpub.com/brain