An Empirical Comparative Study of Cost-Sensitive Classification Algorithms
Ming Yan · 2005
This paper describes a study of different cost-sensitive classification algorithms. The purpose of the study is to analyze the behavior of various cost-sensitive algorithms and how the variations in the induction process affect the total misclassification cost, high cost error amount and total misclassification error amount. For the AdaCost method, this paper analyzes why the cost adjustment factor may cause negative effect on its performance, and implements two modification methods that improve performance substantially.