An adaptive neuro-fuzzy inference system for software development risk assessment in uncertain environments

Abstract

Relevance. The paper addresses the problem of achieving effective risk management in software development projects under conditions of severe uncertainty, ambiguity, and high volatility. Traditional risk management approaches, which rely on rigid quantitative techniques or precise mathematical formulations, often prove ineffective due to limited initial information and the highly subjective nature of expert judgments, highlighting the need for flexible, dynamic, and adaptive decision-making models. Aim of the article. The purpose of the study is to introduce, design, and validate an innovative decision-making framework based on an Adaptive Neuro-Fuzzy Inference System to significantly improve both the accuracy and adaptability of risk assessment and classification in software engineering. Tasks. The research focuses on formalizing software project risk parameters under incomplete data, integrating the data-driven learning capabilities of neural networks with fuzzy logic mechanisms into a unified inference architecture, conducting an empirical evaluation using real-world project datasets, and performing a comparative analysis against conventional risk management techniques. Methods. The proposed framework is developed using the Adaptive Neuro-Fuzzy Inference System architecture, which combines the adaptive learning strengths of neural networks with the uncertainty-handling capabilities of fuzzy logic. The methodology is implemented and validated through empirical simulation and comparative testing against classical analytical techniques using real-world project data. Scientific novelty. The study lies in the design and application of a hybrid neuro-fuzzy model that comprehensively addresses uncertainty, automatically accounts for risk interdependencies, and dynamically adapts to variations in changing project environments, offering a significantly higher precision framework compared to traditional standalone fuzzy or statistical solutions. Practical significance. The proposed Adaptive Neuro-Fuzzy Inference System -based framework is highly applicable within decision-support systems, software project monitoring tools, and automated risk-management modules to provide continuous, resilient, and automated risk classification under dynamically evolving or incomplete data conditions. Results. Empirical evaluation using real-world project datasets demonstrates that the proposed ANFIS model substantially outperforms conventional fuzzy inference systems and analytic hierarchy process (AHP) techniques, attaining a high risk-classification accuracy of ninety-four point seven percent while exhibiting excellent stability across diverse input scenarios. Conclusions. The developed neuro-fuzzy framework is established as a highly precise, flexible, and promising solution for advancing risk management practices in software engineering, successfully overcoming the limitations of traditional methodologies regarding precision, adaptability, and flexibility in volatile environments.