Data and reproducibility materials for "Short-term prediction of cyclist mechanical load for power-targeted adaptive e-bike assistance"

This repository contains the anonymized data and reproducibility materials supporting the study “Short-term prediction of cyclist mechanical load for power-targeted adaptive e-bike assistance”. The dataset comprises second-by-second observations collected during 28 real-world mountain bike (MTB) and electric mountain bike (eMTB) rides performed under four riding conditions. The repository includes the anonymized analysis dataset, data dictionary, Python analysis and figure-generation code, model-performance results, persistence-benchmark results, calibration results, sensitivity-analysis results, and software requirements. These materials enable reproduction of the main descriptive and predictive analyses, including short-term prediction of cyclist relative power (%FTP), leave-one-route-out cross-validation, classification of sustained upper-target mechanical-load excursions, probability calibration assessment, and sensitivity analyses using alternative event definitions. Variables and metadata that could potentially enable identification of the participant or riding locations have been excluded from the publicly available dataset.