An improved technique of single-hidden layer feedforword networks for class-imbalanced data

The framework of Single-hidden Layer Feedforword neural Network with random weights (SLFN) has been shown simple in theory and fast in implementation. However, the decision boundary of SLFN classifier can be affected by the class imbalance of the data. In this work, to deal with the class-imbalanced problem on SLFN, a newly designed algorithm called SW-SLFN is proposed. SW-SLFN connects the synthetic minority oversampling technique (SMOTE) with Weighted-SLFN by an adjustable coefficient called balance factor. Experiments show that SW-SLFN obtains good generalization ability and achieve an improved performance when compared with several other SLFN-based classifiers in imbalance learning.