An Effective Monitoring of Women Reproductive Organ Cancer using Mean based KPCA

Abstract

Data dimensionality reduction yields a compressed low-dimensional structure of a particular high-dimensional dataset. Various dimensionality reduction algorithms have been established to achieve these tasks. Although all these approaches have the same objective, methods to the problem are dissimilar. Here we explored the association between attribute reduction methods and the subsequent classification accuracy for three different areas of cancer as Breast, Ovarian and Cervical Cancer. Principal Components Analysis (PCA) is a traditional technique that offers successive linear estimations to a specific high-dimensional observation. This is why it is one of the entire prevalent procedures for dimensionality reduction. However, its usefulness is restricted by its universal linearity and its inability to solve non-linear problems. To solve the dimensionality reduction problem in nonlinear cases, numerous new methods comprising Kernel Principal Component Analysis (KPCA) has evolved. What KPCA established on nonlinear data may not perform well because the results may vary depending on different kernel functions. To overcome this problem, we proposed a descent algorithm called Mean-KPCA technique based on the mean of Final Data obtained using Gaussian, Exponential and Laplacian Kernel in KPCA. In this paper, subsets of the original attributes computed by PCA, KPCA and Mean-KPCA are compared regarding the classification performance achieved with various machine learning algorithms. We consecutively minimised the size of the attribute sets and inspected the fluctuations in the classification results.