Machine learning-based prediction of non-alcoholic fatty liver disease in Chinese perimenopausal women: a retrospective observational study

This study used a real-world single-centre health examination dataset.

Uniform data preprocessing and hyperparameter tuning were applied across eight machine learning algorithms.

Least absolute shrinkage and selection operator regression was used for feature screening to reduce multicollinearity of input variables, and model interpretability was enhanced with Shapley additive explanations analysis.

This study was a retrospective study, and non-alcoholic fatty liver disease was diagnosed through ultrasound examination.

The definition of perimenopausal status was determined only by chronological age, without clinical manifestations or hormonal testing.