Explainable AI Integration in High-Performance FinTech Systems
Year: 2026
Type: conference-paper
Abstract: The rapid rate of deep learning adoption in FinTech has brought about a so-called black-box dilemma: At the very edges of capabilities, powerful models become more greedy with respect to predicting accuracy, at the expense of interpretability. The call of regulatory requirements, like the EU AI Act (2026) and international financial regulatory practices, already imply a transparent decision- traceability in real- time. This research discusses the application of Integrated Gradients (IG) as a complex, mathematically elegant featureimportance attribution method in high performance FinTech applications. In comparison to classic post-hoc models, such as SHAP or LIME, IG uses a gradient-based calculus to provide its near-instant explanations both in the high-frequency trading context and in detecting fraud itself. We have shown that IGbased integration can be used with sub-millisecond latency and meet strong right to explanation criteria.
Author: Kishore Subramanya Hebbar
Institution: Intercontinental Exchange (United States)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7203662016
