Memetic Computation Based SVM for Cancer Classification

• Main Authors: Nguyen Thi Nga, Nguyen Thi Nga(阮氏娥)
• Other Authors: Shinq-Jen Wu
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/9u3urc

碩士 === 大葉大學 === 電機工程學系 === 100 === Cancer is one of the dreadful diseases found in most of the living being, which is one of the challenging studies for scientist towards 21th century. In cancer diagnosis and treatment, cancer classification plays a very important role. With the advent of DNA microarrays technology, constructing gene expression profiles for different cancer types has already become a promising means for cancer classification. However, it offers a challenge for current machine learning research. Microarray datasets are characterized by high dimension and small sample size. Over-fitting is a major problem due to the high dimension, while the small data size makes it worse. Support vector machine (SVM) is statistical classification algorithm that classifies data by separating two classes with the help of a functional hyper plane. SVM is known for good performance on noisy and high dimensional data such as microarray. One main disadvantage of using SVMs is that the performance of classifier depends on setting of parameters. In this thesis, we do classify cancer using gene expression data with a SVM classifier. A hybrid approach of particle swarm optimization (PSO) and simulated annealing (SA) is proposed to determine proper setting of SVM parameters which can improve the quality of SVM model. Our approach is a combination of methods. The motivation is to bring out an effective classification method for cancer by utilizing the strength of various techniques and compensating for their weaknesses. The proposed approach is tested on six benchmark cancer gene expression data sets, namely, colon, leukemia, lung, ovarian, prostate and breast. The experimental results show that the classification accuracy rates of the proposed method are competitive to that of other existing methods. It can be used as an efficient computational tool for microarray data analysis. Keywords: cancer classification, support vector machine, parameter optimization.