Classification on the Imbalanced Data Stream with Concept Drifts Using a G-means Update Ensemble Approach

• Main Authors: Sin-Kai Wang, 王信凱
• Other Authors: Bi-Ru Dai
• Format: Others
• Language: en_US
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/uhhp44

碩士 === 國立臺灣科技大學 === 資訊工程系 === 104 === Concept drift has become an important issue while analyzing data streams. Further, data streams can also have skewed class distributions, known as class imbalance. Actually, in the real world, it is likely that a data stream simultaneously has multiple concept drifts and an imbalanced class distribution. However, since most research approaches do not consider class imbalance and the concept drift problem at the same time, they probably have a good performance on the overall average accuracy, while the accuracy of the minority class is very poor. To deal with these challenges, this paper proposes a new weighting method which can further improve the accuracy of the minority class on the imbalanced data streams with concept drifts. The experimental results confirm that our method not only achieves an impressive performance on the average accuracy but also improves the accuracy of the minority class on the imbalanced data streams.