A hybrid approach for artificial immune recognition system / Mahmoud Reza Saybani

Mahmoud Reza, Saybani (2016) A hybrid approach for artificial immune recognition system / Mahmoud Reza Saybani. PhD thesis, University of Malaya.

Preview

PDF (Thesis PhD) Download (4Mb) | Preview

Abstract

Various data mining techniques are being used by researchers of different domains to analyze data and extract valuable information from a data set for further use. Among all these techniques, classification is one of the most commonly used tasks in data mining, which is used by many researchers to classify instances into two or more pre-determined classes. The increasing size of data being stored have created the need for computer-based methods for automatic data analysis. Many researchers, who have developed methods and algorithms within the field of artificial intelligence, machine learning and data mining, have addressed extracting useful information from the data. There exist many intelligent tools, which try to learn from the patterns in the data in order to predict classes of new data. One of these tools is Artificial Immune Recognition System (AIRS), which has been used increasingly. Results of AIRS have shown its potential for classification purposes. AIRS is an intelligent classifier offering robust and powerful information processing capabilities and is becoming steadily an effective branch of computational intelligence. Although AIRS has shown excellent results, it still has potentials to perform even better and deserves to be investigated. AIRS uses a linear function to determine the amount of resources that needs to be allocated and this linearity increases the running time and thus reduces the performance of AIRS. Resource competition part of AIRS poses another problem, here, premature memory cells are generated and therefore classification accuracy decreases. Further AIRS uses k-Nearest Neighbor (KNN) as a classifier and that makes it severely vulnerable to the presence of noise, irrelevant features, and the number of attributes. KNN uses majority voting and a drawback of this is that the classes with the more frequent instances tend to dominate the prediction of the new instance. The consequence of using KNN is ultimately reduction of classification accuracy.This dissertation presents the following main contributions with the goal of improving the accuracy and performance of AIRS2. The components of the AIRS2 algorithm that pose problems will be modified. This thesis proposes three new hybrid algorithms: The FRA-AIRS2 algorithm uses fuzzy logic to improve data reduction capability of AIRS2 and to solve the linearity problem associated with resource allocation of AIRS. The RRC-AIRS2 uses the concept of real-world tournament selection mechanism for controlling the population size and improving the classification accuracy. The FSR-AIRS2 is a new hybrid algorithm that incorporates the FRA, RRC, and the SVM into AIRS2 in order to produce a stronger classifier. The proposed algorithms have been tested on a variety of datasets from the UCI machine learning repository. Experimental results on real-world machine learning benchmark data sets have demonstrated the effectiveness of the proposed algorithms.

Actions (For repository staff only : Login required)