In essence, ICET works by invoking C4.5 1000 times (Section 3.5). Fortunately, Quinlan's (1992) implementation of C4.5 is quite fast. Table 6 shows the run-times for the algorithms, using a single-processor Sun Sparc 10. One full experiment takes about one week (roughly 23 minutes for an average run, multiplied by 5 datasets, multiplied by 10 random splits, multiplied by 7 misclassification error costs equals about one week). Since genetic algorithms can easily be executed in parallel, there is substantial room for speed increase with a parallel machine. Each generation consists of 50 individuals, which could be evaluated in parallel, reducing the average run-time to about half a minute.