Hierarchical evolution of linear regressors

We propose an algorithm for function approximation that evolves a set of hierarchical piece-wise linear regressors. The algorithm, named HIRE-Lin, follows the iterative rule learning approach. A genetic algorithm is iteratively called to find a partition of the search space where a linear regressor can accurately fit the objective function. The resulting ruleset performs an approximation to the objective function formed by a hierarchy of locally trained linear regressors. The approach is evaluated in a set of objective functions and compared to other regression techniques. Categories and Subject Descriptors I.2.6 [Learning]: concept learning, knowledge acquisition General Terms Algorithms Keywords Genetic algorithms, machine learning, function approximation, regression

• Type: Conference
• Year: 2008
• Where: GECCO
• Authors: Francesc Teixidó-Navarro, Albert Orriols-Puig, Ester Bernadó-Mansilla
• Keywords: Function Approximation, GECCO 2008, Linear Regressors, Objective Functions, Optimization