Figure 6.
The six plots present accuracy in per cent against the number of selected features. Panel a compares A A L 2 and L P B A 40 using natural and t score methods. Accuracy varies substantially for the natural methods before generally increasing, while the t score methods remain comparatively stable after an initial rise. Panels b through f compare Schaefer, C Atlas and C Atlas plus for 100, 200, 300, 400 and 500 regions, respectively. Schaefer natural decreases markedly at lower feature counts in each panel before recovering as more features are selected. Schaefer t score rises rapidly and then remains relatively stable. C Atlas and C Atlas plus rise rapidly at lower feature counts and generally remain at higher accuracy levels with smaller fluctuations. C Atlas and C Atlas plus include t-score series, while their solid series use S V M coefficient features.Feature ranking; LPBA40 and AAL Atlas versus C-Atlas for different number of regions (100, 200, 300, 400, 500). X-axis: Number of selected features; Y-axis: Accuracy (%). Note that starting from the top 50 features in (c)–(f), the classifier can achieve 90% of ACC
Source: Authors’ own work