Abstract
Early detection of knee osteoarthritis (KOA) is essential to improve treatment outcomes and reduce its long-term impact. However, early diagnosis of KOA (EKOA) remains difficult due to the absence of standardised criteria and the subtle or intermittent nature of early symptoms. This study evaluates the utility of Functional Logistic Regression (FLR) as a robust and efficient classifier that integrates clinical data and biomechanical signals-specifically, ground reaction force (GRF) curves represented through Functional Data Analysis-to support EKOA diagnosis. Fifty-six individuals with knee pain were assessed; 25 were diagnosed with EKOA based on Mahmoudian's clinical criteria, and 31 were classified as healthy. Anthropometric data and pain scores (Visual Analogue Scale, VAS) were collected, and GRF signals in three directions (vertical, anteroposterior, and mediolateral) were recorded during walking trials. A FLR model was developed by combining GRF curves, anthropometric measurements, and VAS scores. The mediolateral GRF component showed the strongest discriminative power. The model that combined biomechanical and clinical variables outperformed models using either data source alone, achieving higher accuracy and sensitivity while maintaining parsimony and robustness. Functional Logistic Regression offers key advantages over classical logistic models based on scalar features and black-box machine learning approaches: it allows the direct use of biomechanical signals without prior discretisation and reduces the risk of overfitting while preserving statistical rigour. These results support FLR as a robust and efficient method for early KOA classification based on combined clinical and biomechanical data.