A kernel-regression framework for gait signal enhancement and inter-joint coordination: a single-subject multi-session methods study.

Accurate characterisation of lower-limb joint kinematics and inter-joint coordination is essential for understanding human gait and its neuromechanical control. This methods-focused single-subject, multi-session case study introduces and evaluates a non-parametric kernel-regression framework for gait signal enhancement and coordination analysis. Two kernel-based estimators, the Nadaraya-Watson estimator and Kernel Ridge Regression, were applied to publicly available lower-limb kinematic data from one healthy adult during treadmill walking, hiking, and running. Validation comprised three experiments: (i) denoising and intra-cycle interpolation of joint-angle trajectories, (ii) long-range forecasting of gait cycles and amplitudes in one-dimensional single-joint and eight-dimensional multi-joint representations, and (iii) predictive mapping of inter-joint relationships to construct coordination networks. Performance was quantified using mean squared error and network-derived node strengths across sessions and locomotor modes. The Nadaraya-Watson estimator performs best when local fidelity and high-dimensional forecasting were prioritised, whereas Kernel Ridge Regression provides smoother reconstructions and more stable inter-joint predictability networks. Error maps and node-strength analyses revealed lower errors for anatomically corresponding contralateral joints and higher strengths for proximal joints than for distal joints. These case specific patterns illustrate how the framework can characterise inter-joint predictive structure. Within the present single-subject treadmill case study, the structured validation across denoising, forecasting, and network analysis showed that the proposed kernel-regression framework can be used as an interpretable methods tool for enhancing signal enhancement and for deriving descriptive, data-driven coordination metrics. Broader biomechanical and clinical generalisation will require replication in multi-subject and clinical studies.