A genetic algorithm for peer-review panel composition

The composition of scientific review panels is a constrained optimization problem in which a finite pool of experts must be distributed among multiple panels while balancing scientific expertise and demographic diversity.

As the number of possible panel configurations grows very rapidly with the number of reviewers, exhaustive searches rapidly become computationally impractical.

In this paper I present a genetic algorithm designed to optimize panel composition for the European Southern Observatory (ESO) proposal evaluation process.

Panel assignments are represented through a chromosome-based encoding.

Candidate solutions are evaluated using a fitness function based on four imbalance indicators: scientific expertise, gender, country affiliation, and professional seniority.

The method is tested using real reviewer data from the ESO proposal handling system.

The results show that the genetic algorithm rapidly identifies panel configurations with substantially lower imbalance than those obtained from random assignments and progressively improves the quality of the overall population.

Beyond producing a single optimized configuration, the approach generates a set of high-quality panel realizations that can subsequently be filtered according to additional operational constraints not explicitly included in the fitness function.

Although developed for ESO, the methodology is general and applicable to a wide range of panel-based peer-review systems.