ATM Transaction Forecasting on Daily transaction data (normal)
28.62SMAPE (%)CL-IL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CL-ILBackbone=TimesNet, Forecast Horizon=14-day, Training Strategy=Instance Contrastive Learning (contrastive pretraining with w=1)2025.12 | 28.62 | -0.11 | |
| WECABackbone=TimesNet, Forecast Horizon=14-day, Training Strategy=Weighted-Contrastive Anomaly-Aware Adaptation2025.12 | 28.7 | -0.03 | |
| NTBackbone=TimesNet, Forecast Horizon=14-day, Training Strategy=Normally trained (trained on normal data only)2025.12 | 28.73 | — | |
| FTBackbone=TimesNet, Forecast Horizon=14-day, Training Strategy=Fine-Tuning (adapted on anomalous samples)2025.12 | 31.5 | 2.77 |