Model Performance
Baseline comparison · Accuracy metrics · Feature importance
মূল ফলাফল — V1001
CPU RMSE (Online)
3.5436%
After online learning
RAM RMSE (Online)
0.9298%
After online learning
CPU R²
0.923
AUC-ROC
0.821
LSTM Anomaly
Baseline তুলনা
CPU Forecast RMSE
RAM Forecast RMSE
PRO-VPS vs Baselines
vs ARIMA
53%
ভালো CPU forecast
vs Standard LSTM
74%
ভালো CPU forecast
vs ARIMA (RAM)
65%
ভালো RAM forecast
Multi-Horizon Prediction
| Horizon | CPU RMSE | CPU R² | RAM RMSE | RAM R² | Status |
|---|---|---|---|---|---|
| 5 মিনিট | 8.43% | 0.484 | 1.06% | 0.639 | ✅ Production |
| 10 মিনিট | 12.70% | -0.17 | 1.35% | 0.412 | গ্রহণযোগ্য |
| 15 মিনিট | 13.44% | -0.31 | 1.48% | 0.291 | ইঙ্গিতমূলক |
| 30 মিনিট | 14.96% | -0.62 | 1.67% | 0.096 | উচ্চ অনিশ্চয়তা |
Decision Distribution — 1113 উইন্ডো
| অবদান | |
|---|---|
| Real KVM VPS production data | ✅ |
| Hybrid PyTorch LSTM + XGBoost | ✅ |
| CPU + RAM proactive forecasting | ✅ |
| Multi-horizon prediction | ✅ |
| Monte Carlo uncertainty | ✅ |
| Online learning + drift detection | ✅ |
| Live production deployment | ✅ |