Fig 1.
A systematic methodology approach from raw IMU signals to DL models evaluation.
Fig 2.
Visualization of TUG test, a data acquisition in FRAILPOL using five IMU sensors mounted on participant’s body limbs.
Table 1.
Overview of GSTRIDE and FRAILPOL datasets.
Fig 3.
IMU data partitioning for GSTRIDE and FRAILPOL datasets using the Sliding Window Technique and Participant-based splitting.
Table 2.
Hyperparameters of CNN algorithm for both GSTRIDE & FRAILPOL datasets.
Table 3.
Hyperparameters of DeepConvLSTM algorithm for both GSTRIDE & FRAILPOL datasets.
Table 4.
Hyperparameters of InceptionTime algorithm for both GSTRIDE & FRAILPOL datasets.
Fig 4.
Training and validation losses of DL algorithms on FRAILPOL dataset: (a) CNN, (b) DeepConvLSTM, (c) InceptionTime; and on GSTRIDE dataset: (d) CNN, (e) DeepConvLSTM, (f) InceptionTime.
Table 5.
Overall classification results of DL algorithms on both GSTRIDE & FRAILPOL datasets across three independent runs (Mean ± Standard Deviation).
Fig 5.
Window-level confusion matrices for (a) (d) CNN (b) (e) DeepConvLSTM, and (c) (f) InceptionTime models evaluated on both datasets: FRAILPOL and GSTRIDE.