A Comparitive Study of Different Data Mining Classification Techniques for Cancer Molecular Pattern Discovery

The most important application of Microarray for gene expression analysis is used to discover or classify the unknown tissue samples with the help of known tissue samples. Several general purpose Data Mining Classification Techniques have been proposed recently and studied to predict/identify the cancer patterns. In this research work, we have focused and studied a few Classification Techniques such as Support Vector Machine (SVM), Nearest Neighbor Classifier (k-NN), ICS4, Non-Parallel Plane Proximal Classifier (NPPC), NPPC-SVM, Margin-based Feature Elimination-SVM (MFE-SVM). The performance of these classifiers in terms of Minimum Threshold Level to predict/identify the Cancer Pattern, Execution Time, Training Time, Memory Usage and Memory Utilization have been analyzed. This research work has applied these Classification Techniques to 10 publicly available datasets, and compared how these Classification Techniques performed in class prediction of test datasets. From our experimental study, it is observed that for different Cancer Patterns, the threshold levels are different to predict the Cancer Pattern by various Classifiers. It is also revealed that the execution time to predict the cancer patterns are different for different Classifiers. That is overall this work has revealed that although it is obvious that Threshold level based Selection method improves both the memory utilization and execution time but finding the best Classifier for Cancer Prediction is still complicated and the performance and efficiency of Classifier in terms of Execution Time and Memory Utilization is vary in each case.