Comparison of statistical power of classifiers, under different metrics for assessing classifier performance. Each dataset included 1000 features per subject, where features were distributed according to a Gaussian distribution within each class. The results are based on k = 1% and n = 150. The classifier performance metric 'average classification accuracy' was compared to the 'AUC statistic', in determining the optimal subset of features after dimensionality reduction. Results are based on 100 simulated datasets.