Extreme Learning Machine-based Classification Algorithm For Uncertain Data

Traditional classification algorithms are widely used on determinate data which have the exact values.However,uncertain data is ubiquitous in many real applications.Uncertain data obeys some distribution.If uncertain data is classified by traditional classification algorithms,it will lead to misclassification.Thus,classifying the uncertain data correctly is of great significance to real applications.Extreme learning machine(ELM)is an optimization algorithm of the single-hidden layer feedforward neural networks.And ELM is also a traditional and powerful classification algorithm.However,ELM can not deal with uncertain data directly.So far some studies have deal with uncertain data with ELM.This kind of algorithms models uncertain data with expected values or sample points and gets the results with voting.The problem is that they will lead to the loss of the distribution information of uncertain data.Furthermore,they do not integrate the uncertain information into the ELM algorithm framework and thus can not deal with data uncertainty well.To tackle these problems,we propose a novel ELM-based uncertain data classification algorithm,called UELM.Firstly,UELM employs exact probability density function(PDF)instead of expected values or sample points to model uncertain data,thus avoiding the loss of uncertain information(probability distribution information of uncertain data).Furthermore,UELM redesigns the traditional ELM algorithm by modifying the received content of input layer and the activation function of hidden layer,thus making the ELM algorithm more applicable to uncertain data.Extensive experimental results on different datasets show that our proposed UELM algorithm outperforms the baselines in accuracy and efficiency.