Title:Hybrid Adaptive Fuzzy Extreme Learning Machine for text classification

Abstract:In traditional ELM and its improved versions suffer from the problems of outliers or noises due to overfitting and imbalance due to distribution. We propose a novel hybrid adaptive fuzzy ELM(HA-FELM), which introduces a fuzzy membership function to the traditional ELM method to deal with the above problems. We define the fuzzy membership function not only basing on the distance between each sample and the center of the class but also the density among samples which based on the quantum harmonic oscillator model. The proposed fuzzy membership function overcomes the shortcoming of the traditional fuzzy membership function and could make itself adjusted according to the specific distribution of different samples adaptively. Experiments show the proposed HA-FELM can produce better performance than SVM, ELM, and RELM in text classification.