An adaptive multiple classifier system based on differential evolution and its application in imbalanced data classification

Imbalanced data exists widely in all domains of our daily life, such as disease diagnosis, mineral resource detection, etc. For the classification of imbalanced data, while ensemble classifiers gave a promising solution for classifying such skewed data, existing ensemble classifiers assume all kinds of imbalanced data share the same characteristics, and a universal solution was carefully designed. However, imbalanced data can be unequable based on its imbalanced ratio, the number of features of the number of examples available for training, so it’s difficult to get good results in all of the data set. In this paper, we propose an adaptive multiple classifier system based on differential evolution algorithm (DE-AMCS), system can choose optimal integration of learning model to complete the classification task. 10 datasets from KEEL are selected toverify the effciency of DE-AMCS, and 5 state-of-the-art imbalanced data classification algorithms are also tested for comparison. Experimental results show that the DE-AMCS is competitive or outperforms the state-of-the-art by using various evaluation metrics as indicators. Finally, DE-AMCS is applied to 5 wells of Jianghan Oil Field. For each well, the precision reaches 100%.