Clinician-Assisted Exploratory Data Analysis Framework for Early Diagnosis of Keratoconus

Advances in corneal imaging have generated vast datasets, making it challenging to extract clinically relevant insights. Machine Learning (ML) provides ophthalmologists with essential tools for keratoconus (KC) detection and clinical decision-making. However, effective Exploratory Data Analysis (EDA) remains crucial for identifying trends, forming hypotheses, and selecting key parameters to build robust ML models. This paper presents a comprehensive EDA framework that fosters collaboration between AI specialists and clinicians to enable early KC detection. Using statistical and visual techniques alongside expert input, a clinical dataset with 79 features from 2,491 cases is pre-processed, analyzed, and effectively prepared for ML modelling. The analysis identified a key subset of features–just 6.3% of the original dataset–which was used to develop and evaluate the classification performance of several ML models. Among them, the Random Forest model achieved 99.6% accuracy, surpassing previous studies and demonstrating the effectiveness of the proposed EDA framework.