📈 Demand Forecasting
Multi-model forecasting with TFT, LightGBM, ARIMA and Conformal Prediction intervals
No data loaded
⚙️ Model Configuration
📊 Forecast Results
🎯 Model Explainability
Understand why the model makes these predictions
📐 Feature Importance (TFT VSN)
Variable Selection Network learns which features matter most.
🧠 Temporal Attention
Which historical days influence each forecast point.
📈 Why This Accuracy?
Historical Patterns
+35%
Weekly Seasonality
+25%
Rolling Statistics
+20%
Trend Component
+12%
External Features
+8%
🎓 Conformal Prediction
Distribution-free coverage guarantee.
P(Y ∈ [L, U]) ≥ 1 - α
Unlike traditional intervals, conformal prediction provides finite-sample guarantees without distributional assumptions.
95.2%Coverage
12.4Avg Width
📋 Performance Metrics
--
RMSE
Root Mean Square Error - penalizes large errors
--
MAE
Mean Absolute Error - average deviation
--
MAPE
Mean Absolute Percentage Error
--
Coverage
% of actuals within prediction interval