Prediction of Hepatitis C Patient by using Support Vector Machine

Abstract

Hepatitis is one of the most often diagnosed infectious diseases worldwide. It is difficult to identify important conclusions from the large amount of data which is accessible in the medical field. With the advancement of technology, data mining methods have established themselves as the most widely used approaches in a variety of fields. Predictive analysis is one of the areas in the medical sector. The goal of this research is to find ways to diagnose the disease using the machine learning techniques for early prediction in hepatitis C patients based on their clinical examination. The dataset was obtained from the UCI Machine Learning repository. The method for selecting features include leveraging on correlation-based feature selection (CFS) with p-value which is to find the significant independent attributes, hence improving the classifier performance. Bilirubin, Albumin, Protime, Fatigue, Malaise, Spiders, Ascites, varices and Histology were found to be the most significant independent attributes. Consequently, this research performed the prediction of hepatitis C using Support Vector Machine on the selected features, comparing the prediction of hepatitis C with and without CFS with p-value feature selection. The performance evaluation of the algorithm was evaluated using accuracy, precision and recall. Experimental result has demonstrated that CFS with p-value for Support Vector Machine were outperforming with the highest accuracy 0.9388 compared to the default SVM.