Comparison of Imbalanced Data Classification Methods

• Main Authors: CHEN,YI-JEN, 陳逸真
• Other Authors: HWANG,YI-TING
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/ee5d82

碩士 === 國立臺北大學 === 統計學系 === 105 === Many practical data have imbalanced features such as fraud detection, medical diagnosis, facial recognition, etc. The common classification methods perform poorly under imbalanced data. Sampling techniques are proposed to improve the performance of the classification. Nevertheless, the overall performance of these sampling techniques implemented to the classification techniques are not evaluated thoroughly. This thesis uses the Monte Carlo simulation to discuss the performance of the classification and various sampling schemes under various imbalanced settings. Many indices are used to assess the performance. Some guidance is given based on the results of simulations.