Machine Learning Approaches for Optimal Parameter Selection for Hepatitis Disease Classification

Background: In the field of medicine, diagnosis is very important issue. In effort to address the issue Machine learning is being utilized for model development for classification and diagnosis purposes. Feature selection (course of dimensionality reduction) is an important component of ML to increase model performance by reducing redundant features which may degrade model performance accuracy. Objective: The goal of this work is to identify among three feature selection approaches which one gives optimal solution (subset of features) that when used for classification provides the best fit performance accuracy and minimized CPU time. Method: using newly created subsets of features we develop model M for hepatitis disease classification for dataset D (xi,yi) where subsets {x1…,xn} contribute to target variable as the original features do. Chi-square, Genetic Algorithm and Principal Component Analysis (PCA) were implemented for feature selection and three classical machine learning algorithms were used for the classification including Naive Bayes, Support Vector Machines and Logistic Regression. Results: The classification performance of the classifiers on the reduced features for Hepatitis disease is estimated using classification accuracy, recall and precision analysis. Experimental result shows the combination of Chi-Square feature selection method and Logistic classifier achieved the best result having 92% accuracy within a desirable CPU time.