Abstract:
Human behavioural health data have increased over the past decade but remain difficult to combine or reuse because sectors describe behavioural events differently. This study aimed to investigate how behavioural health data can be standardised as machine-actionable, interoperable data objects that support exchange across sectors while remaining compatible with routine workflows. Stakeholder consultations and requirements engineering identified 32 requirements, formalised in a traceability matrix and operationalised through eight competency questions. Semantic Web methods were used to develop a standardised behavioural event data object (Atom) as a JSON Schema, classification model ontologies and registries, and persistent identifiers supporting Linked Data projection into knowledge graphs. Evaluation used two weeks of wrist-accelerometer data from eight participants, processed using two classification methods, to demonstrate cross-model data handling. Controlled error seeding and confusion-matrix metrics assessed Atom validation before downstream querying and aggregation. Structural conformance, event-label integrity, duplicate detection, and temporal-gap detection achieved 100% accuracy and F1 scores, with no false positives across both methods. Direct-identifier screening also produced no false positives but showed lower detection performance (F1 scores = 42.42% and 30.85%). Downstream queries successfully retrieved events within specified time windows, summarised behaviour duration, and aggregated classifications into shared behavioural categories through persistent identifiers and knowledge graphs. The resulting data object and supporting resources provide a practical bridge between human behaviour measurement and the Semantic Web, enabling interoperable and auditable behavioural health data across research, clinical, and public health systems. External evaluation is required to assess implementation, scalability, and governance across independent stakeholder systems.
