Human Activity Recognition on PAMAP2
98.6AccuracyMulti-Branch CNN-GRU with attention
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-Branch CNN-GRU with attentionFramework time=10s2026.06 | 98.6 | — | — | — | |
| LSTM-AE (STS method)Framework time=1s, Segmentation type=inter-HA segmentation2026.06 | 98.4 | — | — | — | |
| LSTMFramework time=10s2026.06 | 97.64 | — | — | — | |
| STELLA2026.07 | 96.48 | — | 96.52 | — | |
| Transformer-LSTMFramework time=10s2026.06 | 96.3 | — | — | — | |
| rTsfNet2026.04 | 95.35 | 93.53 | — | — | |
| SenseHAR2026.07 | 95.32 | — | 95.08 | — | |
| TCNet2026.04 | 95.22 | 94.52 | — | — | |
| MA-CNN2026.07 | 95.14 | — | 94.99 | — | |
| Conv-LSTM2026.07 | 94.68 | — | 94.46 | — | |
| MchCnnGRU2026.04 | 94.38 | 91.79 | — | — | |
| mobileHART2026.04 | 93.47 | 90.67 | — | — | |
| MA-DNN2026.07 | 92.16 | — | 92.04 | — | |
| Crossformer2026.04 | 91.84 | 91.29 | — | — | |
| RAG-HAR2025.12 | 91.6 | — | 91.12 | — | |
| CNN2026.07 | 90.4 | — | 90.36 | — | |
| iTransformer2026.04 | 90.18 | 89.46 | — | — | |
| RandomForest2026.04 | 89.92 | 89.39 | — | — | |
| ICGNet2026.04 | 89.39 | 85.36 | — | — | |
| LightTS2026.04 | 89.14 | 89.05 | — | — | |
| Informer2026.04 | 87.32 | 86.25 | — | — | |
| SensorLLM2026.04 | 87.2 | 86.2 | — | — | |
| Pyraformer2026.04 | 87.01 | 86.05 | — | — | |
| Transformer2026.04 | 86.96 | 85.29 | — | — | |
| Reformer2026.04 | 86.81 | 85.22 | — | — | |
| FiLM2026.04 | 86.81 | 86.05 | — | — | |
| TimesNet2026.04 | 86.23 | 85.97 | — | — | |
| DNN2026.07 | 85.82 | — | 86.22 | — | |
| ADFE2025.12 | 85.69 | — | 77.84 | — | |
| CLMMParams (M)=25.48, FLOPs (G)=1.46, Latency (ms)=3.8382026.04 | 85.2 | — | — | — | |
| METIER2025.12 | 83.97 | — | 77.66 | — | |
| FEDformer2026.04 | 83.48 | 81.92 | — | — | |
| Semi-Recurrent CNN-LSTM attention modelFramework time=5s2026.06 | 83.42 | — | — | — | |
| Transformer-like2025.12 | 82.88 | — | 74.84 | — | |
| MC-CNN2025.12 | 79.77 | — | 72.72 | — | |
| DeepConvLSTM2026.04 | 78.2 | 78.4 | — | — | |
| DeepConvLSTM2025.12 | 76.23 | — | 67.54 | — | |
| MASTERParams (M)=21.37, FLOPs (G)=1.79, Latency (ms)=8.3532026.04 | 76.2 | — | — | — | |
| CroSSLParams (M)=17.79, FLOPs (G)=1.68, Latency (ms)=4.1722026.04 | 72.6 | — | — | — | |
| COCOAParams (M)=15.47, FLOPs (G)=0.65, Latency (ms)=2.8342026.04 | 71.5 | — | — | — | |
| COSMOParams (M)=19.83, FLOPs (G)=1.04, Latency (ms)=2.5232026.04 | 69.6 | — | — | — | |
| CMCParams (M)=19.83, FLOPs (G)=0.59, Latency (ms)=2.2382026.04 | 66.7 | — | — | — | |
| STMAEParams (M)=23.44, FLOPs (G)=1.14, Latency (ms)=2.2282026.04 | 61.6 | — | — | — | |
| LLM as Virtual Annotators2025.12 | 56.7 | — | 53.8 | — | |
| AnyMoSetting=Zero-shot2026.05 | 52.6 | — | 41.5 | 78.2 | |
| UniMTSSetting=Zero-shot2026.05 | 47.2 | — | 43.6 | 63.2 | |
| Hargpt2025.12 | 32.11 | — | 31.57 | — | |
| Gemma 4 26BSetting=Zero-shot, Input Format=Plot2026.05 | 19.4 | — | 10.8 | 35.4 | |
| IMU2CLIPSetting=Zero-shot2026.05 | 14 | — | 11.3 | 26.6 | |
| Gemma 4 26BSetting=Zero-shot, Input Format=Text2026.05 | 13.3 | — | 7.4 | 31.1 | |
| ImageBindSetting=Zero-shot2026.05 | 12.8 | — | 7.7 | 15.5 | |
| HARGPTSetting=Zero-shot2026.05 | 11.1 | — | 2.1 | 23 | |
| IMUGPTSetting=Zero-shot2026.05 | 8.9 | — | 1.5 | 19.3 | |
| NormWearSetting=Zero-shot2026.05 | 7.9 | — | 1.6 | 10.5 | |
| FlowFMEvaluation Protocol=Fine-tuning2025.12 | 0.892 | — | 88.77 | — | |
| DiffFMEvaluation Protocol=Fine-tuning2025.12 | 0.8902 | — | 88.77 | — | |
| 1D-Diffusion ClassifierEvaluation Protocol=Fine-tuning2025.12 | 0.8763 | — | 87.68 | — | |
| SENvT-u4Evaluation Protocol=Fine-tuning2025.12 | 0.8488 | — | 84.54 | — | |
| SENvT-u7Evaluation Protocol=Fine-tuning2025.12 | 0.8477 | — | 84.45 | — | |
| FlowFMEvaluation Protocol=Transfer learning2025.12 | 0.8467 | — | 83.73 | — | |
| 1D-DINOEvaluation Protocol=Fine-tuning2025.12 | 0.8341 | — | 82.23 | — | |
| DiffFMEvaluation Protocol=Transfer learning2025.12 | 0.8171 | — | 80.9 | — | |
| 1D-DINOEvaluation Protocol=Transfer learning2025.12 | 0.8111 | — | 79.81 | — | |
| BERTEvaluation Protocol=Fine-tuning2025.12 | 0.7561 | — | 75.27 | — | |
| SENvT-u4Evaluation Protocol=Transfer learning2025.12 | 0.7157 | — | 69.12 | — | |
| SENvT-u7Evaluation Protocol=Transfer learning2025.12 | 0.7098 | — | 68.69 | — | |
| b-LSTM-S2025.12 | — | — | 86.8 | — | |
| CI-BabyMamba-HARChannel Processing=Channel Independent2026.02 | — | 65.46 | — | — | |
| Crossover-BiDir-BabyMambaDirectionality=Bidirectional2026.02 | — | 65.67 | — | — | |
| DeepConvLSTM2026.02 | — | 67.79 | — | — | |
| DeepConvLSTM2026.02 | — | 66.2 | — | — | |
| EfficientKAN (K-M-M)KAN variant=EfficientKAN, Architecture configuration=K-M-M2026.05 | — | 54.7 | — | — | |
| EfficientKAN (M-K-M)KAN variant=EfficientKAN, Architecture configuration=M-K-M2026.05 | — | 44.9 | — | — | |
| EfficientKAN (M-M-K)KAN variant=EfficientKAN, Architecture configuration=M-M-K2026.05 | — | 54.5 | — | — | |
| FastKAN (K-M-M)KAN variant=FastKAN, Architecture configuration=K-M-M2026.05 | — | 48.2 | — | — | |
| FastKAN (M-K-M)KAN variant=FastKAN, Architecture configuration=M-K-M2026.05 | — | 51.2 | — | — | |
| FastKAN (M-M-K)KAN variant=FastKAN, Architecture configuration=M-M-K2026.05 | — | 53.1 | — | — | |
| FourierKAN (K-M-M)KAN variant=FourierKAN, Architecture configuration=K-M-M2026.05 | — | 51.8 | — | — | |
| FourierKAN (M-K-M)KAN variant=FourierKAN, Architecture configuration=M-K-M2026.05 | — | 49.6 | — | — | |
| FourierKAN (M-M-K)KAN variant=FourierKAN, Architecture configuration=M-M-K2026.05 | — | 45.8 | — | — | |
| KAN (K-M-M)KAN variant=KAN, Architecture configuration=K-M-M2026.05 | — | 54.9 | — | — | |
| KAN (M-K-M)KAN variant=KAN, Architecture configuration=M-K-M2026.05 | — | 45 | — | — | |
| KAN (M-M-K)KAN variant=KAN, Architecture configuration=M-M-K2026.05 | — | 54.9 | — | — | |
| KAN-MLP-Mixer2026.05 | — | 56 | — | — | |
| LarctanKAN (K-M-M)KAN variant=LarctanKAN, Architecture configuration=K-M-M2026.05 | — | 55.2 | — | — | |
| LarctanKAN (M-K-M)KAN variant=LarctanKAN, Architecture configuration=M-K-M2026.05 | — | 44 | — | — | |
| LarctanKAN (M-M-K)KAN variant=LarctanKAN, Architecture configuration=M-M-K2026.05 | — | 55.9 | — | — | |
| MLPHAR2026.05 | — | 54 | — | — | |
| Softmax LSTM2025.12 | — | — | 79.2 | — | |
| SSL-WearablesEvaluation Protocol=Transfer learning2025.12 | — | — | 72.5 | — | |
| SSL-WearablesEvaluation Protocol=Fine-tuning2025.12 | — | — | 78.9 | — | |
| TinierHAR2026.02 | — | 74.07 | — | — | |
| TinierHAR2026.02 | — | 74.07 | — | — | |
| TinyHAR2026.02 | — | 73.22 | — | — | |
| TinyHAR2026.02 | — | 73.22 | — | — | |
| Triplet LSTM2025.12 | — | — | 81.1 | — | |
| Triplet LSTMvariant=OTL2025.12 | — | — | 90.4 | — | |
| WavKAN (K-M-M)KAN variant=WavKAN, Architecture configuration=K-M-M2026.05 | — | 56.3 | — | — | |
| WavKAN (M-K-M)KAN variant=WavKAN, Architecture configuration=M-K-M2026.05 | — | 45.2 | — | — | |
| WavKAN (M-M-K)KAN variant=WavKAN, Architecture configuration=M-M-K2026.05 | — | 46.7 | — | — |