Time Series Imputation on ETTm1
0.002MSEMSTN-Transformer
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MSTN-TransformerMask R.=12.5%2025.11 | 0.002 | 0.009 | — | |
| MSTN-BiLSTMMask R.=12.5%2025.11 | 0.003 | 0.011 | — | |
| MSTN-TransformerMask R.=Avg2025.11 | 0.005 | 0.026 | — | |
| MSTN-BiLSTMMask R.=Avg2025.11 | 0.007 | 0.027 | — | |
| Bridge-TS-2Ratio=12.5%2025.12 | 0.016 | 0.084 | — | |
| Non-stationaryRatio=12.5%2025.12 | 0.017 | 0.087 | — | |
| TimesNetRatio=12.5%2025.12 | 0.018 | 0.089 | — | |
| UniTS-STtraining=single-task separately trained from scratch2024.02 | 0.019 | 0.087 | — | |
| Bridge-TS-2Ratio=25.0%2025.12 | 0.019 | 0.092 | — | |
| Non-stationaryRatio=25.0%2025.12 | 0.019 | 0.093 | — | |
| GTMAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.021 | 0.096 | — | |
| TimesNetRatio=25.0%2025.12 | 0.022 | 0.097 | — | |
| T1Point Missing Ratio=Avg2026.02 | 0.022 | 0.091 | — | |
| T1Scenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.022 | 0.091 | — | |
| Bridge-TS-2Ratio=37.5%2025.12 | 0.024 | 0.101 | — | |
| Non-stationaryRatio=37.5%2025.12 | 0.024 | 0.101 | — | |
| TSPulseMasking Strategy=Hybrid, Evaluation Protocol=Finetuned2025.05 | 0.024 | — | — | |
| TimesNettraining=single-task separately trained from scratch2024.02 | 0.027 | 0.107 | — | |
| TimesNetMask Ratio=Avg2024.11 | 0.027 | 0.107 | — | |
| TimesNetRatio=37.5%2025.12 | 0.027 | 0.108 | — | |
| TimesNetSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.027 | 0.107 | — | |
| TimesNetAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.027 | 0.107 | — | |
| GPT4TSMask Ratio=Avg2024.11 | 0.028 | 0.105 | — | |
| GPT2(3)Number of Layers=3, Sequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.028 | 0.105 | — | |
| GPT4TSAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.028 | 0.105 | — | |
| Bridge-TS-2Ratio=50.0%2025.12 | 0.029 | 0.112 | — | |
| Non-stationaryRatio=50.0%2025.12 | 0.029 | 0.112 | — | |
| Peri-midFormerMask Ratio=12.5%2024.11 | 0.03 | 0.108 | — | |
| Peri-midFormerMask Ratio=25%2024.11 | 0.031 | 0.112 | — | |
| TimesNetRatio=50.0%2025.12 | 0.033 | 0.118 | — | |
| Peri-midFormerMask Ratio=37.5%2024.11 | 0.034 | 0.117 | — | |
| Peri-midFormerMask Ratio=Avg2024.11 | 0.034 | 0.116 | — | |
| FEDformerRatio=12.5%2025.12 | 0.034 | 0.131 | — | |
| CSDIPoint Missing Ratio=Avg2026.02 | 0.034 | 0.114 | — | |
| CSDIScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.034 | 0.114 | — | |
| TSPulseMasking Strategy=Block, Evaluation Protocol=Finetuned2025.05 | 0.035 | — | — | |
| Station.training=single-task separately trained from scratch2024.02 | 0.036 | 0.126 | — | |
| Non-stationary TransformerMask Ratio=Avg2024.11 | 0.036 | 0.126 | — | |
| StationarySequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.036 | 0.126 | — | |
| StationaryMask R.=Avg2025.11 | 0.036 | 0.126 | — | |
| PatchTSTScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.038 | 0.119 | — | |
| ZEUSProtocol=Zero-shot, Masking Strategy=Random2026.07 | 0.038 | 0.116 | — | |
| ModernTCNScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.04 | 0.124 | — | |
| HGTS-FormerSequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.04 | 0.126 | — | |
| Peri-midFormerMask Ratio=50%2024.11 | 0.041 | 0.125 | — | |
| TimeMixer++Publication Year=2025, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.041 | 0.127 | — | |
| TimesNetScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.045 | 0.13 | — | |
| PatchTSTMask Ratio=Avg2024.11 | 0.047 | 0.14 | — | |
| PSW-IScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.047 | 0.131 | — | |
| PatchTSTSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.047 | 0.14 | — | |
| PatchTSTAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.047 | 0.14 | — | |
| FEDformerPublication Year=2022b, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.048 | 0.152 | — | |
| TimesNetPublication Year=2023, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.049 | 0.147 | — | |
| LogTranstraining=single-task separately trained from scratch2024.02 | 0.05 | 0.154 | — | |
| ReformerMasking Ratio=Average (12.5%, 25%, 37.5%, 50%), Publication Year=20202026.02 | 0.05 | 0.154 | — | |
| Auto.training=single-task separately trained from scratch2024.02 | 0.051 | 0.15 | — | |
| AutoformerMask Ratio=Avg2024.11 | 0.051 | 0.15 | — | |
| SaitsPoint Missing Ratio=Avg2026.02 | 0.051 | 0.127 | — | |
| SAITSScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.051 | 0.127 | — | |
| AutoformerSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.051 | 0.15 | — | |
| ModernTCNProtocol=Supervised, Masking Strategy=Random2026.07 | 0.051 | 0.152 | — | |
| FEDformerRatio=25.0%2025.12 | 0.052 | 0.162 | — | |
| TimeMixer++Scenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.052 | 0.136 | — | |
| Non-Stationary TransformerMasking Strategy=Hybrid, Evaluation Protocol=Supervised2025.05 | 0.053 | — | — | |
| TimesNetProtocol=Supervised, Masking Strategy=Random2026.07 | 0.053 | 0.152 | — | |
| MAS4TSMasking Ratio=Average (12.5%, 25%, 37.5%, 50%), Publication Year=20242026.02 | 0.054 | 0.147 | — | |
| Re.training=single-task separately trained from scratch2024.02 | 0.055 | 0.166 | — | |
| ReformerMask Ratio=Avg2024.11 | 0.055 | 0.166 | — | |
| PatchTSTMissing ratios=0.5 and 0.72026.02 | 0.055 | 0.153 | — | |
| ReformerSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.055 | 0.166 | — | |
| PWS-IMissing ratios=0.5 and 0.72026.02 | 0.057 | 0.146 | — | |
| TSPulseMasking Strategy=Hybrid, Evaluation Protocol=Zero-shot2025.05 | 0.057 | — | — | |
| DlinearRatio=12.5%2025.12 | 0.058 | 0.164 | — | |
| PatchTSTProtocol=Supervised, Masking Strategy=Random2026.07 | 0.058 | 0.157 | — | |
| MICNPublication Year=2023a, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.059 | 0.17 | — | |
| TimesNetMasking Strategy=Hybrid, Evaluation Protocol=Supervised2025.05 | 0.059 | — | — | |
| FreTSMissing ratios=0.5 and 0.72026.02 | 0.061 | 0.168 | — | |
| FED.training=single-task separately trained from scratch2024.02 | 0.062 | 0.177 | — | |
| FEDformerMask Ratio=Avg2024.11 | 0.062 | 0.177 | — | |
| FEDformerMasking Ratio=Average (12.5%, 25%, 37.5%, 50%), Publication Year=20222026.02 | 0.062 | 0.177 | — | |
| FEDformerSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.062 | 0.177 | — | |
| FedformerAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.062 | 0.177 | — | |
| iTransformerScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.063 | 0.159 | — | |
| ZEUSProtocol=Zero-shot, Masking Strategy=Block2026.07 | 0.064 | 0.142 | — | |
| GPT4TSProtocol=Supervised, Masking Strategy=Random2026.07 | 0.065 | 0.169 | — | |
| HiSNOTMissing ratios=0.5 and 0.72026.02 | 0.066 | 0.15 | — | |
| iTransformerRatio=12.5%2025.12 | 0.067 | 0.181 | — | |
| iTransformerMissing ratios=0.5 and 0.72026.02 | 0.068 | 0.173 | — | |
| BRITSScenario=Point missing, Missing Ratios=0.1, 0.3, 0.5, 0.7 (avg), Sequence Length=962026.02 | 0.07 | 0.166 | — | |
| Linear InterpolationMasking Strategy=Hybrid, Evaluation Protocol=Statistical2025.05 | 0.07 | — | — | |
| In.training=single-task separately trained from scratch2024.02 | 0.071 | 0.188 | — | |
| InformerMask Ratio=Avg2024.11 | 0.071 | 0.188 | — | |
| InformerMasking Ratio=Average (12.5%, 25%, 37.5%, 50%), Publication Year=20212026.02 | 0.071 | 0.188 | — | |
| InformerSequence Length=96, Mask Ratios=12.5%, 25%, 37.5%, 50%2023.02 | 0.071 | 0.188 | — | |
| InformerAveraging=varying data missing ratios, Protocol=point-wise missing ratios2025.02 | 0.071 | 0.188 | — | |
| TimeMixerPublication Year=2024b, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.072 | 0.178 | — | |
| ModernTCNProtocol=Supervised, Masking Strategy=Block2026.07 | 0.074 | 0.181 | — | |
| iTransformerPublication Year=2023, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.075 | 0.177 | — | |
| DLinearPublication Year=2023, Sequence Length=1024, Masking Ratios=12.5%, 25%, 37.5%, 50% (Average)2025.08 | 0.076 | 0.191 | — | |
| Non-Stationary TransformerMasking Strategy=Block, Evaluation Protocol=Supervised2025.05 | 0.076 | — | — |