A Workflow for Geospatial–Temporal Numerical Association Rule Mining from Enriched Cycling Activity Data

Training Center XML (TCX) files store detailed cycling activity data. However, conventional association rule mining methods do not directly capture their geographic context. This study proposes a geospatial–temporal method for numerical association rule mining from enriched TCX-based cycling activity data. The method transforms fused trackpoint-level observations, enriched with OpenStreetMap and Open-Meteo attributes, into a property graph representation of recurring segments, connected paths, and segment runs. Candidate rules are evaluated within graph-derived spatial contexts using a multi-component fitness function that balances rule quality, spatial specificity, path extent, and repeated traversal evidence. The proposed method is implemented in a Julia workflow that supports rule construction, optimization-based search, rule archiving, and map-based interpretation. An evaluation on cycling activities shows that the method generates compact and locally consistent rules linking performance, temporal, infrastructure, and weather attributes. The resulting rule archive preserves both rule quality measures and spatial metadata, enabling the discovered patterns to be inspected in their geographic context.