Human activity recognition using smartphones

1. Clinical question and intended use

A physiological-sensing workflow maps engineered smartphone-signal features to six activity categories. The current page documents the supplied data and network, without presenting performance from an earlier model as a result of this version.

2. Cohort, measurements and endpoint

The source is UCI Human Activity Recognition Using Smartphones. The adapted file contains the six labels LAYING, SITTING, STANDING, WALKING, WALKING_DOWNSTAIRS and WALKING_UPSTAIRS. The current model has 561 inputs, replacing the narrower feature table used by the former article.

Source: Human Activity Recognition Using Smartphones. Dataset license: CC-BY-4.0.

3. Model

The model has 561 encoded inputs and 6 outputs. No architecture-selection experiment is recorded in this project. The diagram shows the topology used by the saved model.

Output values are uncalibrated model scores. Use the output encodings and decision rule documented with this project; do not assume independent sigmoid scores sum to one.

4. Training strategy

The saved training configuration uses QuasiNewton with CrossEntropy.

No separate completed training task is stored in this report. The optimizer configuration is not evidence that a training run completed; no loss or epoch result is reported here.

5. Model selection and baseline

No model selection experiment is recorded for this version. The validation subset guides fitting where a training report is present; it is distinct from the held-out test rows.

6. Clinical validation

The supplied .nd has no saved testing analysis. Its 2059 testing-role records are a configured subset, not an evaluated result. No accuracy, sensitivity, specificity, precision or ROC AUC is available for this current version.

7. Workflow and reproducibility

Use the same input schema and category order when calculating outputs.

Research workflow: eligible record or image → measurement and schema checks → model score → expert review → confirmatory clinical method where appropriate. No model output should be used as an autonomous diagnosis or treatment instruction.

8. Safety, generalizability and governance

The CSV does not include participant identifiers for a subject-grouped audit. The model’s random row roles do not establish new-participant performance, and the adapted numerical values should be checked against the source preprocessing before reproducing normalized HAR benchmarks. Complete and save the training and evaluation tasks before reporting metrics. No external validation is documented. Physiological-signal specialist and expert review are needed to assess subject transfer.

References

• Human Activity Recognition Using Smartphones. Jorge Reyes-Ortiz; Davide Anguita; Alessandro Ghio; Luca Oneto; Xavier Parra. https://doi.org/10.24432/C54S4K
• Dataset terms: Creative Commons Attribution 4.0 International.