Leaderboard

HARENA is a living benchmark for sensor-based human activity recognition, comparing architectures on recognition performance on unseen subjects and measured deployment cost.

Snapshot
Training Runs
10,564
Models
19
Datasets
29
Devices
3

Filters

Datasets
All datasets 29/29
Models
All models 19/19
Building Blocks
All building blocks 7/7

Rankings

Compare recognition performance across models, and how often each model lies on a per-dataset Pareto front on the selected device.

ModelBuilding Blocks
CNN-HAR Affiliations withheld during review
4.44% ±3.20%
CNN
74.4%24.1% 20.7% 41.4% 24.1%
Triple-Cross-Attn Affiliations withheld during review
3.98% ±3.32%
AttnCNNDense
74.0%20.7% 24.1% 27.6% 20.7%
TinyHAR Affiliations withheld during review
3.61% ±3.17%
AttnCNNDenseRNN
73.6%34.5% 31.0% 20.7% 34.5%
TinierHAR Affiliations withheld during review
3.00% ±2.90%
AttnCNNDenseRNN
73.0%41.4% 10.3% 13.8% 37.9%
DANA Affiliations withheld during review
2.55% ±3.40%
CNNDenseRNN
72.5%10.3% 10.3% 10.3% 10.3%
LITEWAY-Light Affiliations withheld during review
2.49% ±2.60%
AttnCNN
72.5%69.0% 55.2% 89.7% 69.0%
Attend-and-Discriminate Affiliations withheld during review
2.48% ±3.17%
AttnCNNDenseRNN
72.5%0.0% 0.0% 0.0% 0.0%
DynamicWHAR Affiliations withheld during review
2.23% ±2.99%
CNNDenseGraph
72.2%0.0% 6.9% 0.0% 0.0%
DeepConvLSTM-Attn Affiliations withheld during review
0.18% ±3.51%
AttnCNNDenseRNN
70.2%0.0% 0.0% 0.0% 0.0%
Random Forest 100 trees, depth=8
-0.02% ±0.04%
FE
70.0%86.2% 86.2% 37.9% 86.2%
MLP-HAR Affiliations withheld during review
-0.57% ±2.43%
DenseSpec
69.4%3.4% 6.9% 0.0% 3.4%
LITEWAY Affiliations withheld during review
-0.59% ±2.70%
AttnCNN
69.4%13.8% 48.3% 31.0% 13.8%
DeepConvLSTM Affiliations withheld during review
-1.74% ±3.27%
CNNDenseRNN
68.3%0.0% 0.0% 0.0% 0.0%
MLP-Mixer Affiliations withheld during review
-2.31% ±4.76%
Dense
67.7%3.4% 24.1% 3.4% 3.4%
Guan-LSTM Affiliations withheld during review
-4.76% ±4.38%
DenseRNN
65.2%0.0% 0.0% 0.0% 0.0%
SA-HAR Affiliations withheld during review
-6.12% ±3.85%
AttnCNNDense
63.9%0.0% 0.0% 0.0% 0.0%
DeepConvShallowLSTM Affiliations withheld during review
-6.86% ±3.69%
CNNDenseRNN
63.1%0.0% 10.3% 0.0% 0.0%
SVM RBF kernel, C=1
-16.85% ±3.43%
FE
53.1%0.0% 0.0% 0.0% 0.0%
k-NN k=5, distance-weighted
-20.42% ±3.37%
FE
49.6%3.4% 3.4% 3.4% 3.4%