Soft Computing in Financial Technology: A Survey of Fuzzy, Evolutionary and Neuro-Fuzzy Methods

Abstract

Financial problems are rarely crisp. Borrower creditworthiness must often be judged under vagueness, financial markets exhibit nonlinear and non-stationary behavior, and portfolio and trading decisions must be made within large, uncertain search spaces. Soft computing, which comprises fuzzy logic, neural networks, evolutionary computation, and swarm intelligence, was conceived to handle exactly this kind of imprecision, and for three decades it has been applied to financial problems. This survey classifies the soft-computing paradigms, reviews their applications across six areas of financial technology (credit scoring, fraud detection, forecasting, portfolio optimization, bankruptcy prediction, and financial statement analytics), and compares them along dimensions that matter in financial practice. The review finds that no single paradigm suits every financial problem; hybrid methods, particularly neuro-fuzzy systems, frequently offer the best balance between predictive accuracy and interpretability. The survey closes by discussing open challenges, including explainability, non-stationarity, class imbalance, and integration with deep learning.