Outlier Detection
Modellus automatically identifies data points that deviate significantly from the regression curve, helping you spot measurement errors or interesting phenomena.
Overview
When you perform a regression on table data, Modellus evaluates each data point against the fitted curve. Points that fall significantly far from the curve are flagged as outliers. These are visually marked in Chart shapes to help you interpret your data.
How Outlier Detection Works
The detection process:
- A regression is computed on the selected data (linear or quadratic)
- For each data point, the residual (distance from the curve) is calculated
- Points with residuals exceeding a statistical threshold are flagged
- The
isOutlierIterationmethod marks these iterations - Charts render outlier points with a distinct visual style
Visual Appearance in Charts
| Point Type | Appearance | Meaning |
|---|---|---|
| Normal | Filled circle, series color | Data point consistent with the regression model |
| Outlier | Different marker style (hollow or contrasting color) | Data point far from the fitted curve |
Interpreting Outliers
An outlier point may indicate:
- Measurement error — The data point was incorrectly recorded
- Experimental anomaly — Something unusual happened during that measurement
- Model limitation — The chosen regression model (linear/quadratic) doesn't capture the full behavior
- Interesting phenomenon — A genuinely different behavior worth investigating
Working with Outliers
Excluding outliers from regression
If you determine that certain points are measurement errors, you can exclude them by selecting a narrower range of cells in the Table before applying regression. Only the selected cells participate in the curve fitting.
Comparing regressions
Try both linear and quadratic regression on the same data. If the linear model produces many outliers but the quadratic model doesn't, the relationship may be non-linear — consider upgrading your theoretical model.
Technical Details
The outlier detection is performed per-iteration by the calculation engine's isOutlierIteration method. The Chart shape queries this for each Y-series data point during rendering, applying the outlier visual style to flagged iterations.