Explainable Deep Learning Models for Trustworthy Decision Support in High-Stakes Data Science Applications

Nurul Adha Oktarini, Saputri and Adeen, Amjad and Aleena, Jamil (2026) Explainable Deep Learning Models for Trustworthy Decision Support in High-Stakes Data Science Applications. Journal of Data Science, 2026 (02). pp. 15-37. ISSN 2805-5160

	Text 852 - Published Version Available under License Creative Commons Attribution. Download (36kB)
	Text jods2026_02.pdf - Published Version Available under License Creative Commons Attribution. Download (343kB)

Abstract

Deep learning models are increasingly deployed in high-stakes domains such as healthcare, finance, and public decision systems, where predictive errors and opaque reasoning can lead to significant societal consequences. Despite their superior predictive capabilities, most deep learning systems remain black-box models, limiting transparency, regulatory compliance, and user trust. Existing explainable artificial intelligence (XAI) approaches often function as post-hoc add-ons and rarely integrate explanation stability into the model optimization process. To address this gap, this study proposes a unified explainable deep learning framework that embeds model-agnostic and model-specific interpretability techniques directly into a multi-objective optimization pipeline. The framework jointly optimizes predictive performance, computational efficiency, and explanation stability under predefined deployability constraints. Experiments were conducted on benchmark datasets representing high-stakes risk assessment and resource allocation scenarios using MLP and attention-based architectures. Results show that explainability-integrated models achieved a stability score of 0.89 (vs. 0.72 baseline) and reduced representation shift by 39%, while maintaining competitive predictive performance (ROC-AUC up to 0.901, <1.2% degradation). Human-centered evaluation further demonstrated a significant increase in trust scores (4.18 vs. 3.12, p < 0.001). These findings indicate that embedding explainability as a structural design principle enhances robustness and trustworthiness without sacrificing accuracy. The study contributes a deployable framework for responsible AI in high-stakes decision support systems

Item Type:	Article
Uncontrolled Keywords:	Explainable AI; Deep Learning; Trustworthy AI; Decision Support Systems; Model Interpretability
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software
Depositing User:	Unnamed user with email masilah.mansor@newinti.edu.my
Date Deposited:	26 Feb 2026 03:26
Last Modified:	26 Feb 2026 03:53
URI:	http://eprints.intimal.edu.my/id/eprint/2299

Actions (login required)

View Item