A Novel Classification Method for Class-imbalanced Data and Its Application in microRNA Recognition

Abstract

For non-coding RNA gene mining, especially microRNA mining, there are many challenges in the classification of imbalanced data. A novel classification method based on the Adaboost algorithm is proposed to handle the imbalance of positive and negative cases. Unstable-Adaboost is improved with respect to the initial weight assignment, the base classifier selection, the weight adjustment mechanism and other aspects. Furthermore, the Stable-Adaboost algorithm is proposed, which adjusts the weight of the sample set to rapidly achieve a more balanced training set. In addition, the Stable-Adaboost algorithm can ensure that the follow-up training set is maintained in a balanced state by optimizing the weight adjustment mechanism of incorrectly classified samples and stabilizing the classification performance. Experimental results show the superiority of Unstable-Adaboost and Stable-Adaboost in imbalance classification.

Keywords

• Non-coding RNA
• Class imbalance
• Ensemble learning
• Adaboost algorithm