This section examines the baseline performance of the algorithms. In Section 4.1.1, we look at the average cost of classification of the five algorithms on the five datasets. Averaged across the five datasets, ICET has the lowest average cost. In Section 4.1.2, we study test expenditures and error rates as functions of the penalty for misclassification errors. Of the five algorithms studied here, only ICET adjusts its test expenditures and error rates as functions of the penalty for misclassification errors. The other four algorithms ignore the penalty for misclassification errors. ICET behaves as one would expect, increasing test expenditures and decreasing error rates as the penalty for misclassification errors rises. In Section 4.1.3, we examine the execution time of the algorithms. ICET requires 23 minutes on average on a single-processor Sparc 10. Since ICET is inherently parallel, there is significant room for speed increase on a parallel machine.