A self-adaptive evolutionary weighted extreme learning machine for binary imbalance learning

It is known that the problem of imbalanced data sets widely exists in various application fields. The weighted extreme learning machine (WELM) was proposed. It solved the L 2 -regularized weighted least squares problem in order to avoid the generation of an over-fitting model and to obtain better classification performances in imbalanced data sets when compared with extreme learning machine. While a WELM algorithm can address class imbalance issues, the random assignment of input parameters and the training sample weights generated according to class distribution of training data have been found to affect the performance of WELM. The aim of this study was to propose a self-adaptive differential evolutionary weighted extreme learning machine (SDE-WELM) which utilized a self-adaptive differential evolutionary to find the optimal input weights, hidden node parameters, and training sample weights of the WELM and exploit an appropriate criterion to be used as the fitness function for binary imbalance learning. The experimental results of the majority of the 40 data sets examined in this study indicated that the proposed method had the ability to achieve a better classification performance when compared with a weighted extreme learning machine (WELM), ensemble weighted extreme learning machine, evolutionary weighted extreme learning machine, and an artificial bee colony optimization-based weighted extreme learning machine and the four popular ensemble methods which combine data sampling and the Bagging or Boosting used in support vector machine as base classifier.