Federated LSTM-LRP Networks with SMOTE and LIME for Interpretable Gait Classification

Gait analysis is essential for the diagnosis and monitoring of various biomechanical and neurological conditions. Clinical reliability is compromised due to imbalanced data and lack of transparency in existing models. This study uses vertical GRF features of patients to analyze gait on the publicly available GaitRec dataset. The proposed approach addresses the limitations of traditional models using Borderline SMOTE to balance the dataset and FedAvg-SVRG to stabilize gradient updates across dispersed nodes and enable federated learning with stable convergence without centralizing data. Enhanced LRPs are combined with LSTM to improve transparency in pattern recognition, and LIME offers instance-level interpretability. A deep learning model is used to improve the accuracy of gait classification, achieving 99.72% on single run execution and 99.45±0.14% for 10-fold cross-validation. The proposed deep learning framework holds significant potential for robust gait classification and clinical decision making.