Walking speed modulates tibialis anterior but not vertical ground reaction force: A multimodal SPM and machine learning analysis – Data and code

This OSF project accompanies the manuscript: "Neuromuscular and kinetic adaptations to walking speed: A multimodal paired SPM, explainable machine learning, and equivalence testing study" (submitted for publication). We used paired Statistical Parametric Mapping (SPM) with cluster‑based sign‑permutation tests (α = 0.05, 5000 permutations) and a Random Forest classifier with leave‑one‑subject‑out (LOSO) cross‑validation and SHAP to compare multi‑axial ground reaction forces (Fz, Fy, Fx) and surface EMG from six lower‑limb muscles between slow (0.5 m/s) and fast (1.0 m/s) treadmill walking in 58 healthy young adults. Functional principal component analysis (fPCA) extracted waveform features, and equivalence testing (TOST) was performed on vertical GRF. ## Corrected Key Findings - GRF: No statistically significant clusters for any force axis (paired SPM, p > 0.05). Fz differences were spatially diffuse under equivalence testing. - Tibialis anterior (TA): Nine significant clusters (Cohen’s dz = 0.67–0.96). - Additional muscles (gastrocnemius medial and lateral, vastus lateralis, rectus femoris, semitendinosus): 1–8 significant clusters each. - Machine learning: Random Forest with fPCA features classified walking speed with 78.0% LOSO accuracy; TA PC1 was the most important predictor (mean |SHAP| = 0.143). ## Contents - **Code/** – Full pipeline: preprocessing, multi‑axial SPM, fPCA, LOSO Random Forest, SHAP, equivalence testing, and figure generation. - **Data/processed/** – Time‑normalised waveform RDS files (3‑axis GRF, 6‑muscle EMG) ready for analysis. - **outputs/figures/** – Figures 1–5 and S1–S2 as 300 dpi TIFFs. - **outputs/tables/** – SPM cluster tables, SHAP importance, confusion matrix, three‑speed results. ## Reproducibility All analyses were conducted in R (version 4.6.0). See the README for step‑by‑step instructions. The original dataset is from Katmah et al. (2026), available on PhysioNet (DOI: https://doi.org/10.13026/r0ea-7161). ## License Code: MIT | Processed data: original PhysioNet license | Figures: CC‑BY 4.0 ## Contact For questions, open an issue on this OSF project or email nyantakyi.appiah@gmail.com.