Abstract
Accurate and timely detection of postural control impairments in stroke patients is crucial for effective rehabilitation and fall prevention. Traditional clinical assessments often rely on qualitative observation and handcrafted features, which may fail to capture the nonlinear and uncertain nature of postural deficits. This study addresses these limitations by introducing a hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) with Type-2 fuzzy logic activation to robustly classify sensory dysfunction under altered balance conditions. Using an EquiTest-derived dataset of 8316 labeled samples from 700 participants across six standardized sensory manipulation scenarios, the proposed method achieved 97% accuracy, 96% precision, 97% sensitivity, and 96% specificity, outperforming conventional CNN and other baseline classifiers. The approach demonstrated resilience to measurement noise down to 1 dB SNR, confirming its robustness in realistic clinical environments. These results suggest that the proposed system can serve as a practical, non-invasive tool for clinical diagnosis and personalized rehabilitation planning, supporting data-driven decision-making in stroke care.