| < Day Day Up > |
|
We have shown that certain medical diagnosis problems, in which it is important to lower the total cost of Type I and Type II errors, could be treated as cost minimization problems. We used the cost matrices to obtain cost trade offs and incorporated these trade offs into a linear GA-based classification system. After incorporation of the cost preferences, the resulting classification system (ICPB-GA) performed better than classification systems that do not incorporate the cost preferences (LDA and GA).
In our research, we assumed that the Type I and Type II error costs are constant. In certain medical situations these costs may vary overtime and may follow a statistical distribution. The current approach can be easily modified to incorporate costs that are not constant. It is likely that in the event costs are not constant, the performance of ICPB-GA could be impacted by the distribution of the costs of Type I and Type II errors. Future research should focus on investigating the impact of different cost distributions on the performance of ICPB-GA.
| < Day Day Up > |
|