An experimental study of Multi-Objective Evolutionary Algorithms for balancing interpretability and accuracy in fuzzy rulebase classifiers for financial prediction

Abstract

This paper examines the advantages of simple models over more complex ones for financial prediction. This premise is examined using a genetic fuzzy framework. The interpretability of fuzzy systems is oftentimes put forward as a unique advantageous feature, sometimes to justify effort associated with using fuzzy classifiers instead of alternatives that can be more readily implemented using existing tools. Here we investigate if model interpretability can provide further benefits by realizing useful properties in computationally intelligent systems for financial modeling. We test an approach for learning momentum based strategies that predict price movements of the Bombay Stock Exchange (BSE). The paper contributes an experimental evaluation of the relationship between the predictive capability and interpretability of fuzzy rule based systems obtained using Multi-Objective Evolutionary Algorithms (MOEA).