Beyond the Black Box: A Review of Quantitative Metrics for Neural Network Interpretability and Their Practical Implications

Abstract

As neural networks continue to grow in complexity and find applications in critical domains such as healthcare, finance, and autonomous systems, the demand for transparent and trustworthy AI has never been greater. This paper provides a comprehensive review of quantitative metrics used to evaluate the interpretability of neural networks, focusing on key measures—fidelity, complexity, robustness, and sensitivity—and examining their respective advantages, limitations, and suitability across different model architectures. In addition, the review explores major challenges in interpretability assessment, including data quality, bias, scalability, and generalizability, while highlighting emerging approaches such as causal and interactive interpretability. By addressing these core issues and advancements, the paper aims to bridge the gap between high model performance and meaningful transparency, ultimately contributing to the development of more accountable and trustworthy AI-driven decision-making systems.