Robust nonparametric curve smoothing that resists outliers using the Epanechnikov kernel and weighted median
Try Interactive DashboardThe weighted median ignores outlier magnitude - only position in sorted order matters. Achieves up to 40% lower error than mean-based smoothers on contaminated data.
The Epanechnikov kernel is optimal in the mean squared error sense among all second-order kernels. Provides efficient, bounded support weighting.
Explore the algorithm with real-time parameter controls, animations, and side-by-side comparisons of median vs mean smoothing.
| Metric | EPA Median | EPA Mean | Improvement |
|---|---|---|---|
| RMSE (8% outliers) | 0.108 | 0.184 | 41% better |
| MAE (8% outliers) | 0.082 | 0.127 | 35% better |
| Breakdown Point | 50% | 0% | Infinitely more robust |