Canadian Journal of Pure & Applied Sciences

Abstract: The work presented in this paper describes hybrid approaches that employ principal component analysis (PCA) and multiple discriminant analysis (MDA) methods for microarray classification. The paper first describes a hybrid approach that incorporates PCA and Fisher linear discriminant analysis (FDA) for microarray classification. This hybrid approach effectively solves the singular scatter matrix problem caused by small training samples. To increase the effective dimension of the projected subspace the use of MDA instead of FDA is explored. The performance of the system is evaluated by projecting data to several subspaces incrementally. The resulting incremental hybrid system improves the accuracy of classification. The paper discusses a comprehensive evaluation of the hybrid systems. The hybrid systems were tested on a dataset of 62 samples (40 colon tumor and 22 normal colon tissues). The results show that the use of incremental hybrid system increased the accuracy of classification of microarray data which will lead to better diagnosis of cancer and other diseases.