การปรับค่าเสียหายในการจำแนกประเภทผิดเพื่อการเรียนรู้ที่อ่อนไหวต่อค่าเสียหายบนชุดข้อมูลชีวการแพทย์ที่ไม่สมดุล

Abstract

The data where number of examples in each class differ significantly or imbalance data can be found in various application domains. The traditional supervised learning following the Empirical Risk Minimisation principle, which minimises the misclassification regardless of the types of error, often yields a classification model that generalises poorly on the minority class. Cost-sensitive learning is one of the promising approaches to introducing inductive bias into the model for imbalance data classification. This thesis the aim comparative study of misclassification cost and initial weight assignment strategies for AdaBoost. And bring about to propose method for automatically determine suitably cost of misclassification and initial weight. In this thesis, we studied three strategies for determining misclassification costs for an imbalance dataset and incorporated such costs into a cost- sensitive AdaBoost algorithm. The strategies consist of Imbalance Ratio which is determine misclassification cost from ratio of each class instance, Grid Search which is find expected parameter procedure for learning step and Distribution Correction that is modify the initial weight by sample size in those target class. Apply whole strategies with Cost-Sensitive AdaBoost. The experimental results based on five imbalance biomedical testbeds. The results are appear the imbalance ratio strategy seemed to over- estimate the misclassification costs and as a result yielded a model which is too focused on the minority class. The grid search improved upon the traditional AdaBoost on some datasets but is still comparable to AdaBoost overall. And the distribution correction strategy seemed to outperform all other strategies. It is therefore recommended that the proposed distribution correction method is the most effective strategy in terms of imbalance-aware performance measures.