Research And Application Of Classification Algorithm Based On Extreme Learning Machine

Classification is an important task of artificial intelligence,which plays an important role in the recognition and cognition of information.The goal of this proj ect is to solve the data problems with imbalanced distribution effectively by digging the important information in the training sample and building a reasonable training model.Many practical application,e.g.fraud identification and intrusion detection,face the critical problem of imbalanced data.Therefore,it is of great significance and importance to solve the data problems with imbalanced distribution.The algorithm proposed in this paper is based on the hierarchical framework of classification and single-level framework of classification integrates five important modules:target class selection,strategy of weight updating,base classifier,winner-take-all and deletion mechanism,which can solve the imbalanced learning effectively.The contributions of this article are as follows:1.A hierarchical framework of classification algorithm is proposed.This algorithm can improve recognition accuracy of target class in the current level and reduce the confusion between the target and non-target classes,through the hierarchical framework and five important modules in the single-level framework of classification.In a word,the proposed algorithm can solve the problem with data with imbalanced distribution effectively.2.The selection algorithm of target class is proposed in this paper.The selection of target class is based on G-means,which can select the class with the highest accuracy and recall rate as target class.The design of this function is important for the training model and decision-making step,and is one of the influencing factors of the recognition ability of the single-level framework of classification.3.Strategy of weight updating based on target class is proposed in this paper.Compared with the strategy of weight updating in AdaBoost algorithm,this algorithm focuses on improving the weights of samples belonging to the target class and the classes that are easily confused with the target class.The strategy of weight updating proposed in this paper has the advantage of adaptively updating sample weights,which achieves the complementarity with the hierarchical framework of classification algorithm.4.This paper proposes winner-take-all and delete mechanism.This algorithm proposed in this paper gets final result of single-layer framework of classification by using the strategy of winner-take-all instead of traditional voting mechanism,which can ensure the highest recognition accuracy of target class.After obtaining the final result of single-level,deletion mechanism is run to delete training samples belonging to target class,which can reduce the confusion between target class and non-target classes.The strategy of winner-take-all and deletion mechanism can be regarded as the decision mechanism of single-level framework of classification.In order to verify the effectiveness of the proposed algorithm,the KEEL database,breast cancer,net attacks data sets are used in experiments.Compared with the experimental results of other algorithm,the proposed algorithm gets the better recognition results in majority of data sets,which can better solve the problem of data with imbalanced distribution.