Imbalanced Classification Methods Based On Extreme Learning Machine And The Application

Imbalanced classification is an important topic in the field of data mining and machine learning,which has attracted more and more attention and becomes a new challenge for the academic and industry communities.The imbalanced learning problem concernes on the performance of learning algorithms with the insufficient data or the skewed data in each class.Due to the inherent complex characteristics of imbalanced data sets,learning from such data requires new algorithms and tools to transform vast amounts of raw data efficiently into information and knowledge.Extreme learning machine algorithm has become a popular classification method due to its excellent performance and faster computational efficiency.However,it has some inherent flaws due to its simplely setting parameters.How to improve the classification performance of the existing algorithms,and fit for more practical classification problems is the urgent problems need to be solved.This paper studies the imbalanced classification problem from the algorithm level on the basis of the extreme learning machine algorithm.The main researches of this paper is as follows:(1)Aiming at the instability performance caused by random initialization parameters of extreme learning machine algorithm,an improved weighted extreme learning machine algorithm with brain storm optimization algorithm is proposed.Brain storm optimization algorithm is adopted to optimize the hidden layer parameters of the weighted extreme learning machine.The experimental results show that the proposed method can effectively improve the classification accuracy of the imbalanced data by the weighted extreme learning machine.(2)Aiming at the network structure of extreme learning machine,an adaptive CCR-ELM algorithm with variable-dimension BSO algorithm is proposed.Not only the hidden layer parameters and trade-off parameters,but also the number of hidden layer nodes are optimized.Because of the variable number of hidden layer nodes,and the individual length of the population will be change during the evolution process.As a result of this,a variable-dimensional BSO algorithm was proposed to find the current optimal CCR-ELM network structure.Experiments show that the proposed algorithm has the stable performance,and weak sensitivity to the imbalanced rate.(3)Aiming at the imbalanced classification problem with few labeled data,a trans-fer weighted extreme learning machine algorithm is proposed.A transfer learning strategy is adopted to realize the knowledge transferring from the source domain to the target domain,and the unlabeled target domain data is used as guide samples to construct a new classifier to classify imbalanced target data.The experiments indicate that the proposed algorithm has better and more stable classification performance than the imbalanced classification algorithms and the other transfer extreme learning machine algorithms,and it inherits the efficiency of the extreme learning machine algorithm.The proposed algorithm is applied to fault diagnosis of coal belts,and the effectiveness of the algorithm in practical applications is verified.