From raw to refined: the influence of preprocessing on ML performance for liver disease detection

Background A computer-based method is needed for non-invasive detection of chronic liver diseases, which are often asymptomatic, progressive, and potentially fatal. A good prediction model enhances the understanding of liver disease using Machine Learning (ML) methods to address this widespread health issue. Data preprocessing is vital in ML classification as it cleans and standardizes the data, which improves the model performance addressing the issues like missing values, feature scaling, and categorical encoding; this in turn minimizes the risk of biased or incorrect predictions. Few studies have systematically quantified how different preprocessing steps affect model performance in liver disease prediction. Methods This study is among the first to comprehensively evaluate the effect of normalization, imputation methods, and feature encoding on multiple ML models for liver disease prediction. We systematically compare preprocessing pipelines (scaling, imputation, feature encoding) across five common machine learning classifiers. We introduce a preprocessing sensitivity index (PSI) to quantify model performance variance due to preprocessing changes. Results We show that certain preprocessing strategies consistently outperform others in liver disease datasets—insights not previously reported. After pre-processing, it is found that there is a significant increase in prediction accuracy; the increment has been found to be over 8–12% after-data preprocessing. Conclusion It is found that the preprocessing not only helps in mitigating issues such as data inconsistencies and noise but also enhances the model’s ability to detect subtle patterns within the data.