Metrics for measuring error extents of machine learning classifiers