An Empirical Analysis of PCA-SVM Model for Cancer Microarray Data Classification

Cancer is nowadays becoming the main reason for increasing the death toll worldwide. The death percentage is increasing due to the unavailability of a proper method to identify cancer at an early stage. Machine learning plays a vital role in deploying an automated model which will help in better diagnosis. The conventional cancer diagnosis is fully based on biopsy data which misses the important aspects of cancer such as proliferation rate, certain behavior of tissues. Hence to indemnify the cancer subtypes, genetic data may play an important role. The microarray data contain the genetic information of a patient with a huge amount of genetic information as compared to the number of samples. To overcome this limitation, the dimensionality reduction technique needs to be applied before applying any kind of ML algorithm for classification. In this paper, we have used the principal component analysis (PCA) as a dimensionality reduction technique. After that supervised machine learning algorithm, SVM is used as the classifier due to its efficiency in high-dimensional datasets and kernel trick. The above-said methodology has been applied to four cancer datasets such as breast cancer, prostate cancer, colon cancer, cervical cancer. Finally, the performance of PCA-SVM and SVM without emphasizing the usage of gene selection technique before the classification of any microarray data with different kernels has been compared in terms of accuracy, precision, and specificity.