ML Model Performance
Best Model Accuracy
96.8%
Ensemble LSTM
Training Time
43 min
-15 min faster
Data Points
2.4M
+180K new records
Model Drift
2.3%
Within tolerance
🤖 Active ML Models
Ensemble LSTM
Deep Learning
Champion
96.8%
MAPE:
3.2%
RMSE:
145.2
MAE:
89.4
Training Time:
43 min
XGBoost Regressor
Gradient Boosting
Challenger
94.2%
MAPE:
5.8%
RMSE:
198.7
MAE:
124.3
Training Time:
12 min
Prophet
Time Series
Active
91.5%
MAPE:
8.5%
RMSE:
234.1
MAE:
156.8
Training Time:
8 min
Transformer
Attention Model
Training
67%
Epochs:
134/200
Loss:
0.0234
Val Loss:
0.0289
ETA:
22 min
Model Accuracy Comparison
Training Performance Over Time
🔍 Model Insights & Feature Importance
Top Features (Ensemble LSTM)
Historical Demand (30d)
0.92
Seasonality Index
0.78
Economic Indicators
0.65
Marketing Spend
0.54
Weather Data
0.43
Model Diagnostics
Overfitting Risk
Low
Train/Val loss ratio: 1.06 (healthy)
Data Drift
Medium
2.3% drift detected in last 30 days
Prediction Confidence
High
92.5% average confidence score
Bias Detection
Low
No significant bias across categories
🔄 Training Queue & Experiments
Neural Prophet v2.1
Advanced time series with external regressors
Processing
67% - 22 min remaining
Multi-Task LSTM
Joint forecasting for correlated products
Queued
Position: #2
Causal Impact Model
Promotional effect quantification
Queued
Position: #3