Efficacious Discriminant Analysis (Classifier) Measures for End Users

Table 9

This table compares Abdou’s results [16] using estimated misclassification cost (EMC) and as measures. The highlighted values indicate the best results. EMC is a cost, and thus the lower the value, the better; WOE is best. estimates the net impact; the higher the number, the better; is best. The results are consistent with other studies; generally, AI-derived algorithms outperform manual algorithms.

Original and reanalysis results
Test	EMC	at ratio+ =
	5 : 1	ratio+ = 1	ratio+ = 1.5	ratio+ = 2.1	ratio+ = 2.5
	0.4627	0.3484	0.4223	0.4764	0.5068
PA	0.6232	0.4377	0.5299	0.5975	0.6354
	0.5679	0.4315	0.5555	0.5896	0.6259
	0.6964	0.4590	0.5561	0.6271	0.6670