Extreme Learning Machine Based Diagnosis Models for Erythemato-Squamous Diseases

摘要

Extreme learning machine based features selection algorithms are proposed in this paper for diagnosing erythemato-squamous diseases. The algorithms adopt the traditional ELM (extreme learning machine), EM-ELM (the error minimum extreme learning machine) and K-ELM (kernel extreme learning machine), respectively, to evaluate the power of the detected feature subset. The improved F-score and SFS (sequential forward search) strategy are combined to detect feature subsets. To detect a much more accurate diagnosis model for erythemato-squamous diseases, an ensemble diagnosis model is constructed by combining three models (classifiers) built on three feature subsets detected by proposed feature selection algorithms respectively. 5-fold cross validation experiments are conducted to test the performance of each feature selection algorithm, and the ensemble model. Experimental results demonstrate that the ensemble model has got the best accuracy. Its highest and average classification accuracy in 5-fold cross validation experiments are 100% and 98.31%, respectively.