Please use this identifier to cite or link to this item:
|Title:||Universal machine learning classifier using extreme learning machines||Authors:||Chen, Jinnan||Keywords:||DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition||Issue Date:||2019||Abstract:||Classification is one of the most essential tasks in machine learning which could be applied to many areas and solve practical problems. Also, it can give us some accurate prediction based on the currently collected data and then automatically realize the objective of trend prediction, pattern recognition or segmentation like the development of autonomous vehicles, or the stock market forecasting. It has gained lots of focus in the modern society and has become part of people’ daily life, so we can see the broad application in the industry. Generally, we will face three types of recognition tasks, binary, multi-class classification, as well as multi-label classification with the difference in the input data features. The general way to address this problem is applying back propagation algorithms into each of the three kinds of problems, but it could be time-consuming and computationally expensive. In order to make our classifier more intelligent to be capable of handling all types of classification datasets regardless of the difference in label types and speed up our training time, we aimed to develop a universal classifier. It could automatically identify the classification type all these classification tasks with high speed and good performance based on the application of extreme learning machine, a powerful algorithm which could significantly improve the training process without losing the accuracy. By using different scales of datasets, we could see the better performance of the proposed classifier and compared with the results of other popular algorithms, so the advantage of the ELM-based universal classifier could be demonstrated. In order to get a universal classifier, we proposed a method to identify the classification type in advance through detecting the feature of the training samples and ensemble the algorithms together to make it a more robust universal classifier which could handle more sophisticated datasets. Through tuning the hyperparameters of the proposed model, we plotted the figures to find the correlation between the performance and the optimum hyperparameters. We also applied the extreme learning machine with some modification as the core part into our algorithm and use several evaluation indexes to give an objective and comprehensive analysis of the proposed method.||URI:||http://hdl.handle.net/10356/77668||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Student Reports (FYP/IA/PA/PI)|
Page view(s) 554
checked on Oct 26, 2020
checked on Oct 26, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.