Fig. 5
Mobilise-D dataset flow. Initially, 75% of the participants were utilized to fine-tune the model for the development of a gait detection model. A fivefold cross-validation approach was employed. During each fold, data was divided into 80% for training and 20% for validation. The model underwent training for 30 epochs, incorporating an early-stop mechanism that halted the training process if the loss on the set did not decrease for 5 consecutive epochs. Model performance evaluation was conducted on the validation set, and the final performance was calculated as the average across the 5 folds. This entire process was repeated three times for three different divisions of the cross-validation folds. The average of these three runs was then calculated. This approach allowed us to evaluate various model configurations and select the most effective one. Ultimately, the best configuration was applied to the 25% test set, comprising 21 participants, and compared against state-of-the-art models.
