Diagnosis of Liver Disease using Cost-Sensitive Support Vector Machine Classifier

In this paper, a reliable decision support system for the accurate diagnosis of liver disease is presented. A benchmark liver disease dataset was utilized to build the prediction model. However, the dataset in question is imbalanced, leading to a biased prediction towards the majority class. A standard error-driven Support Vector Machine (SVM) classifier fails completely to distinguish the diseased patients from the healthy ones due to the large class imbalance present. To alleviate the problem, the SVM classifier is modified into a cost-sensitive design where unequal costs are assigned to different classes. The minority class instances are strongly penalized compared to the majority class. This shifts the decision boundary away from the majority class, leading to a more balanced predictive performance. A grid search technique was utilized to identify the appropriate cost settings to achieve optimal performance. A g-mean score of 65.82%, with a sensitivity of 72.36% and specificity of 59.88% was achieved using our proposed approach. On the contrary, standard error-driven SVM classifier provided a g-mean score of zero. Thus, our proposed methodology has the potential to be an effective tool in the prediction of liver disease.