Comparative Predictive Modeling of Dry Eye Disease: An Integrated Approach Using Decision Tree and Random Forest Techniques

Abstract: Dry Eye Disease (DED) is a multifactorial ocular disorder characterized by tear film instability and ocular surface inflammation, manifesting as discomfort and visual disturbances. Traditional diagnostic methods rely on subjective clinical evaluation and costly procedures, limiting accessibility. This work proposes a machine learning-based, non-invasive approach for predicting DED risk using patient demographics, lifestyle, and reported symptoms. Both Decision Tree and Random Forest classifiers are compared: Random Forest achieves superior accuracy (72.8\%) and F1- score (0.76). Feature importance ranks symptomology and behavioural factors as key predictors, supporting practical early intervention strategies.

Keywords: dry eye disease, machine learning, decision tree, random forest, predictive modelling, feature importance