Developing Predictive Models for Rare Diseases Clinical Trials

Research on rare disorders, which are mostly inherited and often exhibit neurological symptoms, is gravely underrepresented. This research uses gait analysis to categorize patients with hereditary cerebellar ataxia versus healthy controls, and itbuilds a predictive modelling framework for therapeutic trials targeting uncommon disorders. This study presents a gaitclassification methodology for distinguishing individuals with hereditary cerebellar ataxia from healthy subjects, leveraging aninertial measurement unit (IMU) to capture gait metrics and trunk acceleration from thirty diagnosed individuals and one hundredhealthy participants. The dataset undergoes preprocessing steps, including outlier correction using interquartile range (IQR). TheRandom Forest (RF) classifier ranks features by importance, revealing significant predictors like CV-step-length, sLLEAP, andHRAP. To address class imbalance, various data augmentation techniques—under-sampling, over-sampling, SMOTE, GAN, andCT-GAN—are employed. Among these, CT-GAN exhibits the highest effectiveness in producing balanced data, contributing toenhanced classification accuracy. Among a performance parameter utilised to assess an effectiveness of a model, CT-GAN achieved a 90% accuracy rate, an 88% recall rate, an 88% F1-score, and a 0.90 ROC-AUC. The results underscore the advantages of CTGAN in improving model performance for imbalanced clinical data, especially in rare disease prediction