<![CDATA[Given the increasing prevalence and significant health burden of liver diseases globally, improving the accuracy of predictive models is essential for early diagnosis and effective treatment. The purpose of the study is to systematically analyze how different feature selection methods impact the performance of various machine learning classifiers for liver disease prediction. The research method involved evaluating four distinct feature selection techniques—regular, analysis of variance (ANOVA), univariate, and model-based on a suite of classifiers, including decision forest, decision tree, support vector classifier, multi-layer perceptron, and linear discriminant analysis. The result revealed a significant and variable impact of feature selection on model accuracy. Notably, the ANOVA method paired with the multi-layer perceptron achieved the highest accuracy of 0.801724, while the univariate method was optimal for the decision forest classifier (0.741379). In contrast, model-based selection often degraded performance, particularly for the decision tree classifier, likely due to the introduction of noise and overfitting. The support vector classifier, however, demonstrated robust and consistent accuracy across all selection techniques. These findings underscore that there is no universally superior feature selection method; instead, optimal predictive performance hinges on tailoring the selection technique to the specific machine learning model. This study contributes practical, evidence-based insights into the critical interplay between feature selection and model choice in medical data analysis, offering a guide for improving classification accuracy in liver disease prediction. Future work should explore more sophisticated and hybrid feature selection methods to enhance model performance further.]]>
Copyrights © 2025
