Research and Application of Feature Extraction and Multi-objective Machine Learning

A feature extraction and multi-objective machine learning algorithm is proposed based on multi-objective coevolutionary algorithm.Training samples core attributes are found by feature extraction and attribute groups are composed of core attributes and non-core attributes,so the classified accuracy is improved.All attribute groups are supervised clustering by attribute similarity and class tags.The number and center of class families can automatically determined by using the fitness function in machine learning as the goal;in this way,they can avoid the effect of subjective factors and the two key elements owning optimization nature are guaranteed.The class tag using the nearest neighbor method determines a genus of the unclassified samples.At last,the algorithm is demonstrated by the UCI data sets of Liver Disorders,Hepatitis data sets and summery abnormal megathermal forecast in the north area of Zhejiang province.The experiment results indicate that feature extraction and multi-objective machine learning algorithm is better than the well-known NBC,C4.5 and SVM.