A Novel Approach of Disease Diagnostic Prediction Using SMOTE Ensemble Classification

Hepatitis is a liver inflammation caused primarily by viral infection. There are several methods that help with its diagnosis. However, the available datasets for Hepatitis research contain missing values that are imputed by median values. The Synthetic Minority Oversampling Technique (SMOTE) oversamples the minority class, while feature selection algorithms extract critical features from the actual data. One significant disadvantage of existing approaches is their inability to efficiently deal with missing data, resulting in lower statistical effectiveness. In this paper, a novel approach for predicting disease diagnostic is proposed. Firstly, recursive feature elimination (RFE) is used to remove features from the total number declared, followed by univariate analysis, which provides a score to each feature. The principal component analysis (PCA) compresses data. The mutual information score and ReliefF are two automatic feature selection approaches that restore key features. The synthetic minority oversampling technique (SMOTE) oversamples the minority class to obtain balanced data. Several experiments have been conducted using the hepatitis dataset and the framework has shown to predict the disease much more accurately than existing methods.