Human activity recognition leaderboard

RealWorld HAR: macro-F1

Name: RealWorld HAR macro-F1 leaderboard: human activity recognition
Creator: ChatPPG

4 reported results for human activity recognition on RealWorld HAR, ranked by macro-F1. Top reported: SSL ResNet, UK Biobank (Yuan) (0.792). Every row cites…

This page ranks every result ChatPPG has recorded for human activity recognition on RealWorld HAR, measured by Macro-F1 on HAR activities (higher is better). Each number is copied from the publication cited in its row, and ranks apply only within this dataset and metric.

Human activity recognition (IMU): RealWorld HAR, Macro-F1 on HAR activities. Reported rankings; evaluation protocols may differ.
Rank	Method & evaluation	Macro-F1 on HAR activities↑ Higher is better
1	SSL ResNet, UK Biobank (Yuan) Self-supervised ResNet pre-trained on UK Biobank, all layers fine-tuned Evaluation details Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Reported as of: 2026-10-16	0.792
2	Random forest (Yuan) Random forest on hand-crafted features Evaluation details Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Reported as of: 2026-10-16	0.731
3	ResNet from scratch (Yuan) ResNet-V2 trained from scratch Evaluation details Wrist accelerometer resampled to 30 Hz, 10 s windows; subject-wise 5-fold CV (leave-one-subject-out when fewer than 10 subjects); subject-wise macro F1 (Table 2). Reported as of: 2026-10-16	0.705
4	ActiNet actinet-ssl-harnet-hmm-activity-intensity-sci-rep-2026 (ActiNet) Evaluation details Deployed CAPTURE-24 model, wrist ACC; RF+HMM 0.531.	0.661

Rows are ranked only within this task × dataset × metric. Protocols can differ; read each method's evaluation details before comparing. Latest reported row: 2026-10-16.

About the dataset

RealWorld HAR: multi-position smartphone/smartwatch IMU (n=15)

Subjects: 15
Signals: accel, gyro, magnetometer, audio, gps, light
Licence: Not stated

Sources on this board

ActiNet: an open-source tool for activity intensity classification of wrist-worn accelerometry using self-supervised deep learning (paper)

Related leaderboards

Human activity recognition on other datasets

PAMAP2: macro-F1 (7)
opportunity: macro-F1 (6)
capture-24: macro-F1 (5)
capture-24: Cohen's kappa (4)
mhealth: F1 (4)
mhealth: macro-F1 (4)
PAMAP2: F1 (4)
WISDM AR v1.1: accuracy (4)
UCI Human Activity Recognition Using Smartphones (UCI-HAR): F1 (3)
USC-HAD: F1 (3)
DEO: balanced accuracy (1)
har-6bench-avg: macro-F1 (1)

All human activity recognition leaderboards

Data and submissions

Download every leaderboard as JSON or CSV. Missing a published result? Email hello@chatppg.com with the paper link and the exact table it comes from.

Cite this

Cite the page URL and the date you accessed it. Please also cite the original publications listed above.

APA

ChatPPG. (2026). RealWorld HAR macro-F1 leaderboard: human activity recognition [Data set]. ChatPPG Research Hub. https://chatppg.com/research/leaderboards/har/realworld-har-macro-f1-har

BibTeX

@misc{chatppg_har_realworld_har_macro_f1_har,
  author       = {{ChatPPG}},
  title        = {{RealWorld HAR macro-F1 leaderboard: human activity recognition}},
  year         = {2026},
  month        = oct,
  howpublished = {\url{https://chatppg.com/research/leaderboards/har/realworld-har-macro-f1-har}},
  note         = {ChatPPG Research Hub}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this resource, please cite it as below."
type: dataset
title: "RealWorld HAR macro-F1 leaderboard: human activity recognition"
authors:
  - name: "ChatPPG"
url: "https://chatppg.com/research/leaderboards/har/realworld-har-macro-f1-har"
date-released: "2026-10-16"