Kernel Dimensionality Reduction evaluation on various dimensions of effective subspaces for cancer patient survival analysis

In this research, we have extended the use of Kernel Dimensionality Reduction (KDR) in the context of semi supervised learning in particular for micro-array DNA clustering application. We have proposed a new model call K-means-KDR for survival analysis which we aimed to improve the genes classification performance and study the dimension of effective subspaces in cancer patient survival analysis. KDR method was extended and combined with the K-means clustering technique, Cox's proportional hazards regression model and log rank test where KDR contributes in gene classification to determine subgroups from the patient's group. Results from the experiments have indicated that our model has outperformed Support Vector Machines (SVM) in gene classification. We also observed that the best value for dimension of effective subspaces (K) for microarray DNA data is between 10%-20% of the total patients.