Voting Based Extreme Learning Machine Essay examples

562 Words3 Pages
Real valued classification is a popular decision making problem, having wide practical application in various fields. Extreme Learning Machine (ELM) pro- posed by Huang et al.[1], is an effective machine learning technique for real valued classification. ELMis a single hidden layer feedfo5 rward network in which the weights between input and hidden layer are initialized randomly. ELM uses analytical approach to compute weights between hidden and output layer [2], which makes it faster compared to other gradient based classifiers ([3, 4]). Various variants of ELM were recently proposed, which includes Incremental Extreme 10 Learning Machine [5], Kernelized Extreme Learning Machine [6], Weighted Extreme Learning Machine(WELM) [7],…show more content…
are some of the complex valued classifiers designed for real valued classification problems. CCELM out- performs other complex valued classifiers for real valued classification problems. It also performs well when dataset is imbalanced. 35 It has been observed that many practical classification problems have imbalanced data sets[23, 24]. If we classify such data, most of the classifiers favours the majority class due to which most of the instances belonging to minority class are misclassified. To deal with such dataset, various sampling approaches [25] as well as algorithmic approaches are used. Sampling approaches includes over 40 sampling and undersampling techniques. Oversampling replicates a fraction of minority samples while undersampling approach reduces a fraction of majority samples to make dataset balanced. But there is problem with sampling approaches. Oversampling [26] increases redundancy of data and undersampling results in loss of information. In algorithmic approach, classifier design 45 encompasses the measures to handle class imbalance. Most of the neural network based classifiers like FCRBF [4, 3],CCELM[9] minimizes least square error to find optimal weights. Recently proposed WELM minimizes weighted least square error function to find optimal weights between hidden and output layer. In this classifier, residuals of minority
Open Document