Research on Unbalanced Data Classification Based on Hybrid Method

• Main Authors: Nai-Nan ZHANG, 張乃楠
• Other Authors: Ting Ying Chien
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/sy353q

碩士 === 元智大學 === 資訊工程學系 === 107 === Unbalanced data are ubiquitous in real-world datasets. In this paper, we investigate unbalanced data distribution for binary classification, i.e., where the number of majority class instances is significantly greater than the number of minority class instances. It is assumed that traditional machine learning algorithms attempt to minimize empirical risk factors, and, as a result, the classification accuracy of the minority is often sacrificed. However, people are often interested in the minority. Various data-level methods, such as over- and under-sampling, and algorithm-level methods, such as ensemble, cost-sensitive, and one-class learning, have been proposed to improve classifier performance with an unbalanced data distribution. Based on such methods, we proposed a hybrid approach to deal with unbalanced data problem that comprises data preprocessing, clustering, data balancing, model building, and ensemble.