Fig. 1: How FEAT works.
From: A flexible symbolic regression method for constructing interpretable clinical prediction models
a Steps in the genetic programming process. Candidate models are initialized in a population; the best models (parents) are selected via epsilon-lexicase selection; offspring are created by applying variation operations to the parents; and then parents and offspring compete in a survival step using NSGA-II [22]. The process then repeats. b The evaluation of a candidate models’ complexity and performance in Pareto Optimization framework in the Survival step. c Example model in which input features are transformed by logical functions with or without threshold operators.
