Mixture Regression Estimation based on Extreme Learning Machine

Recently, Extreme Learning Machine(ELM) has been a promising tool in solving a large range of regression applications. However, to our best knowledge, there are very few researches applying ELM to estimate mixture regression model. To improve the estimation performance, this paper extends the classical ELM to the scenario of mixture regression. First, based on the idea of fuzzy clustering, a set of fuzzy factors are introduced in ELM to measure the degree of membership for a specific class. Furthermore, a new regularization problem is constructed and then the optimal fuzzy factors can be calculated after multiple iterations. Experiments conducted on toy regression data and a structural response prediction data set show the effectiveness of the proposed algorithm compared to the Support Vector Machine-based algorithm in terms of estimation accuracy and computational cost. Recently, Extreme Learning Machine(ELM) has been a promising tool in solving a large range of regression applications. However, to our best knowledge, there are very few researches applying ELM to estimate mixture regression model. To improve the estimation performance, this paper extends the classical ELM to the scenario of mixture regression. First, based on the idea of fuzzy clustering, a set of fuzzy factors are introduced in ELM to measure the degree of membership for a specific class. Furthermore, a new regularization problem is constructed and then the optimal fuzzy factors can be calculated after multiple iterations. Experiments conducted on toy regression data and a structural response prediction data set show the effectiveness of the proposed algorithm compared to the Support Vector Machine-based algorithm in terms of estimation accuracy and computational cost.