An Explainable Hybrid Deep Learning Framework for Optimized Risk Assessment in Early-Stage Software Development using Design Pattern Features
International Journal of Drug Delivery Technology · 23.05.2026 · 16 (31s)
DOI: 10.25258/ijddt.16.31s.6 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 23.05.2026. Verfasst von Ritesh Kumar und 1 weiteren Autor:innen.
Abstract (Originalsprache)
Early-stage software risk assessment remains a formidable challenge in the software development lifecycle (SDLC), as traditional predictive models typically rely on post-hoc code metrics that are unavailable during the architectural phase. This paper proposes X-HDRAP, a novel Explainable Hybrid Deep Learning Framework designed to identify potential project risks by analyzing Software Design Patterns as primary structural features. Our architecture integrates Convolutional Neural Networks (CNNs) for spatial pattern coupling analysis, Long Short-Term Memory (LSTM) networks for temporal risk propagation, and Recurrent Neural Networks (RNNs) to capture sequential logic dependencies. To address the "black-box" nature of deep learning, we incorporate an interpretability layer using SHapley Additive exPlanations (SHAP), providing developers with transparent, feature-level justifications for risk alerts. The framework was rigorously evaluated against a curated dataset of 500 open-source projects containing over 12,000 validated pattern instances. Empirical results demonstrate that X-HDRAP achieves a 34% improvement in F1-score over state-of-the-art baseline models and successfully detects 73% of architectural risks at least 30 days before their manifestation in the codebase. This research contributes a dual-purpose tool that not only optimizes risk prediction accuracy but also enhances stakeholder trust through explainable AI (XAI) insights.
Zitieren
APA
Kumar, R., & Aggarwal, G. (2026). An Explainable Hybrid Deep Learning Framework for Optimized Risk Assessment in Early-Stage Software Development using Design Pattern Features. International Journal of Drug Delivery Technology, 16(31s). https://doi.org/10.25258/ijddt.16.31s.6
Vancouver
Kumar R, Aggarwal G. An Explainable Hybrid Deep Learning Framework for Optimized Risk Assessment in Early-Stage Software Development using Design Pattern Features. International Journal of Drug Delivery Technology. 2026;16(31s). doi:10.25258/ijddt.16.31s.6
BibTeX
@article{Kumar2026,
author = {Kumar, Ritesh and Aggarwal, Gaurav},
title = {An Explainable Hybrid Deep Learning Framework for Optimized Risk Assessment in Early-Stage Software Development using Design Pattern Features},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {31s},
doi = {10.25258/ijddt.16.31s.6}
}
EUVIMED