Leave-one-out cross-validation provides clarity in radial velocity measurements of exoplanetary systems

Advanced statistical methods help clarify planetary signals hidden within precise radial velocity data. By systematically testing how individual measurements influence a model, researchers have refined the mass estimate of HD 119130 b, clarified the orbital architecture of the Gliese 4276 system, and confirmed the three-planet structure orbiting Gliese 357 by testing for predictive consistency.

Star system: Gliese 357 | Distance from the Sun: 30.8 light years

Radial velocity, or the measurement of a star's "wobble" caused by the gravitational tug of orbiting planets, is a cornerstone of exoplanet detection. As instruments become more precise, they can detect smaller, lower-amplitude signals. However, these tiny shifts are often buried in noise or influenced by individual data points, making it difficult to distinguish between genuine signals and statistical anomalies. To address this, a technique called leave-one-out cross-validation is used to test how sensitive a model is to specific observations.

This method works by repeatedly removing one measurement at a time from a data set and checking how the resulting model performs without it. If removing a specific data point causes a significant change in the outcome, that point is likely driving the model's conclusion. This provides a way to verify the robustness of inferred planetary properties, such as mass or orbital configuration, by ensuring that the results are not dictated by just one or two potentially misleading observations.

Applying this technique to known systems has yielded new clarity. In the case of HD 119130 b, an initial mass estimate appeared anomalously high; the analysis revealed this was driven by only two specific measurements. Removing them significantly lowered the mass and resolved discrepancies between different observational data sets. For the Gliese 4276 system, the method favors a single planet with an eccentric orbit over a two-planet model. Furthermore, in the Gliese 357 system, the method found no predictive evidence for a potential fourth planet, confirming that the system is better described by a three-planet architecture.

Key findings
The high mass originally inferred for HD 119130 b was traced to the influence of only two data points.
Leave-one-out cross-validation identifies that the Gliese 4276 system is best described by an eccentric single-planet model.
The method provides no support for a proposed fourth planet in the Gliese 357 system, reinforcing the previously reported three-planet configuration.
Systematic removal of individual data points helps align model predictions with observed stellar motion.

Scientific study

Enhancing the Interpretability of Radial Velocity Analysis Using Leave-one-out Cross-validation (external link)