A Study of Monotonic One-class Extreme Learning Machine

Abstract

Classification technology has become very popular since it can help make predictions through the characteristics and relevance of data Among them one-class classification is often used for outlier detection or for the data that is imbalance between positive and negative classes Monotonic classification problems mean that there exists monotonicity relation between some features of input data and the decision label Extreme learning machine (ELM) is one of the most commonly used network models in the classification problem By the characteristic of randomly determining the weight ELM can greatly reduce the computation time However because of the existence of training error ELM is not a good model for monotonic classification problems In the past the monotonic classification was often carried out by adding monotonic constraints but this would increase the complexity of the operation Therefore this study hopes to meets the monotonic constraints by adjust the network weight structure of ELM This study satisfied the monotonicity constraints by altering the expression of the network weights so that ELM can handle the monotonic data Furthermore we generate a decision function to handle the problem of one-class classification Through experiments we find that the model we proposed has good stability in dealing with monotonic data classification problems

Date of Award	2020
Original language	English
Supervisor	Sheng-Tun Li (Supervisor)