MoveSmart: A multi-modal dataset for ground reaction force estimation using Apple Watch and force plate data

This dataset contains synchronised inertial sensor data from Apple Watch devices (wrist and waist, ≈100 Hz) and laboratory force plate ground reaction force (GRF) data (1000 Hz) collected from 10 healthy adults performing five activities: walking, jogging, running, heel drops, and step drops. A total of 757 validated trials are included, with 584 trials having complete trial ID matching enabling a validated 3-phase transfer learning analysis. The dataset enables research on wearable-based GRF estimation, biomechanical signal transfer, sensor placement effects, and machine learning model development. The validated 3-phase analysis demonstrates: (1) Waist→Force Plate baseline mapping (mean r = 0.550 ± 0.170), (2) Wrist→Waist transfer validation (mean r = 0.568 ± 0.234), and (3) Wrist→Force Plate deployment testing (mean r = 0.486 ± 0.243), with an excellent transfer learning gap of 0.038 across all activities. Overall deployment readiness is 56.8% (trials with r > 0.5), with running and jogging showing best performance. Data includes synchronised time-series, event metadata, trial matching manifest, quality flags, biomechanical validation results, and full analysis code. This dataset establishes a benchmark reference for wearable-based GRF estimation research and supports reproducibility and open science in biomechanics.