Ensemble Calculator
Combine up to 10 models · averaging · voting · stacking concept · Fleiss’ κ
Models
Max 10 models. Accuracy 0–1 (e.g., 0.85).
Method & weights
Weights will be normalized automatically.
Sample predictions
Enter class labels (e.g., A/B) or probabilities (0–1). For voting use discrete labels.
Results
—
Ensemble prediction
—
Expected accuracy gain
—
Agreement rate
—
Fleiss' κ
Model agreement per sample
Diversity & returns
When ensembles help most: diverse models (low agreement) with moderate accuracy. Below: simulated diminishing returns as more models are added (based on avg diversity).
Method comparison
Simple avg — equal weight, good for probabilities.
Weighted — use accuracy or custom weights.
Majority — hard voting, robust to outliers.
Stacking — meta‑learner combines outputs (concept).
Stacking concept: uses weighted average as meta‑prediction (demo).