Clustering-Based Support Vector Machine (SVM) for Symptomatic Knee Osteoarthritis Severity Classification

The application of machine learning in gait biomechanics of knee osteoarthritis (OA) had been reported in many previous studies. Instead of using radiographic severity classification to grade OA, this study employed pain scores and gait features to characterize knee OA. Pain scores from Knee Injury and Osteoarthritis Outcome Score (KOOS) and Measure of Intermittent and Constant Osteoarthritis Pain (ICOAP) with correlated spatiotemporal, kinematic, and electromyography features were utilized as features. Using k-means clustering to cluster severity, the severity class were then trained and tested using Support Vector Machine classifier. The best performance classification model with an accuracy 85.2% for the training set and 91.2% for the testing set was the dataset of ICOAP constant pain with its correlated gait features, with medium Gaussian Kernel and three severity levels as clustered by K-means. The ranking of features for each dataset was also discovered using the ReliefF algorithm.