PMP-SVM: A Hybrid Approach for effective Cancer Diagnosis using Feature Selection and Optimization

Abstract

Cancer disease is becoming a prominent factor in increasing the death ration over the world due to the late diagnosis. Machine Learning (ML) is playing a vital role in providing computer aided diagnosis models for early diagnosis of cancer. For the diagnosis process the microarray data has its own place. Microarray data contain the genetic information of a patient with a large number of dimensions such as genes with a small sample such as patient details. If the microarray is directly taken without reducing the dimension as the input to any ML model for classification, then Small Sample Size is the resulting issue. So, size of the microarray data needs to be reduces by using either of dimensionality reduction technique or the feature selection technique to increase the model’s performance. In this work, proposed a hybrid model using Principal Component Analysis (PCA), Maximum Relevance Minimum Redundancy (MRMR), Particle Swarm Optimization (PSO) and  Support Vector Machine (SVM) for cancer diagnosis. PCA and MRMR is used for feature selection and PSO is applied to get the optimized feature set. Finally, SVM is applied as the classification model. The proposed model is evaluated against multiple cancer microarray datasets to measure the performance in terms of accuracy, precision, recall, and F1 score. Result shows that proposed model performs better than existing state of art model.