Evolutionary Extreme Learning Machine Weighted Fuzzy-Rough Nearest-Neighbour Classification

Due to the mechanism of the instance-based classification, the feature significance plays an important role in the nearest-neighbour classification tasks. The existence of the irrelevant features would degrade the performance of these algorithms by mischoosing the nearest neighbours. However, these irrelevant features are normally inevitable in the real applications. In this paper, the evolutionary extreme learning machine (E-ELM) algorithm is employed to distinguish the feature significance for the fuzzy-rough nearest-neighbour (FRNN) method. This hybrid learning approach, entitled evolutionary extreme learning machine weighted fuzzy-rough nearest-neighbour algorithm, extracts the feature significance by integrating the parameters from the parallel input node to the output node in E-ELM. Such feature significance is transformed to implement a weighted FRNN method to perform the classification tasks. Systematic experimental results, for both dimensionality reduction and classification problems, demonstrate that the proposed approach generally outperform many state-of-the-art machine learning techniques.