The Role Of Metaheuristics In Supervised Learning Algorithms
The Role Of Metaheuristics In Supervised Learning Algorithms

Summary/Abstract: Many real-world problems, including problems in economics and finance, are difficult to solve analytically (if not impossible), because they may have objective functions that are non-linear, non-continuous, non-differentiable, multi-dimensional, noisy, with many local minima, and have non-linear constraints. Metaheuristics can be used in an attempt to find approximate solutions to such challenging optimization problems. The aim of this paper is to explore the potential of metaheuristics to replace Backpropagation as a training method for feedforward neural networks (multilayer perceptrons). The main characteristics of several metaheuristics are presented and experimental evidence is provided that they perform very well as supervised learners for certain neural network architectures. For comparing the predictive performances of different models evolved by the metaheuristics under consideration, a challenging forecasting application of risky asset prices on capital market is also evaluated.

• Details
• Contents