| Summary: | 碩士 === 輔仁大學 === 管理學研究所 === 96 === In this research we use discriminant analysis, logistic regression, BPN, MARS, and CART to construct the survival model of breast cancer patients. We also use 10-fold cross validation and compare the five models to get the best one.
The results show that if we don’t consider the misplaced cost, the discriminant rate is about 92%~94% except for discriminant analysis(88%). And, the discriminant rate of the BPN model is the best.
If considering the misplaced cost, the results show that BPN model is still the best one when the misplaced-cost ratio of type I error and type II error are 1:1 and 1:2. And when the misplaced-cost ratio is large than or the same as 1:3, the CART model is the best one we got.
|
|---|
