Automated Design of Genetic Programming Classification Algorithms for Financial Forecasting Using Evolutionary Algorithms

Abstract

In this work two metaheuristic algorithms namely, a genetic algorithm (GA) and grammatical evolution (GE) are used to configure genetic programming classification algorithms for financial forecasting. The performance of the classifiers evolved through a GA and GE design are compared to the performance of classifiers evolved using the traditional manual design approach. Fifteen stocks from varied sectors are selected to evaluate the performance. Additionally, the fitness landscape of the design space evolved by grammatical evolution and the genetic algorithm is evaluated. Results demonstrate that GE designed algorithms evolve classifiers that perform better than those designed by a GA and manually designed. Furthermore, it is established that the GA design space is more rugged than the GE design space.