EUVIMEDEUVIMED entdecken
Crossref

ADRL-CostNet: A Deep Reinforcement Learning Framework for Continuous and Self-Adaptive Software Cost Estimation in Agile Environments

Debanjan Ghosh, Arvind Kumar Pandey

International Journal of Drug Delivery Technology · 29.05.2026 · 16 (35s)

DOI: 10.25258/ijddt.16.35s.88 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 29.05.2026. Verfasst von Debanjan Ghosh und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Static prediction models producing one estimate at the start of a project have frequently been employed to estimate software expenses. In Agile settings, where demands, team performance, backlog volume, and technical debt constantly evolve over time, such techniques are sometimes insufficient. Addressing this problem, this work suggests an Adaptive Deep Reinforcement Learning–based Cost Estimation Framework (ADRL-CEF), which views the software development process as a changing and developing system. Based on present project circumstances, the framework employs a learning agent that revises cost-to-complete predictions over sprints. The estimating process considers key factors including sprint velocity variability, requirement volatility, defect density, backlog progress, and technical debt. Deep Reinforcement Learning approaches, including Deep Q-Network (DQN) and Advantage Actor–Critic (A2C) are used to carry out the agent, which then enhances its forecasts via repetitive learning directed by accuracy-based input. Real-world Agile project information and a controlled synthetic dataset were used to assess the suggested method. The outcomes reveal that ADRL-CEF outperforms conventional regression and machine learning models, reaching around an 18–32% reduction in mean relative error (MRE). These results imply that in agile environments reinforcement learning offers a more adaptable and flexible solution for cost estimating. The framework generally encourages ongoing estimation and provides a useful strategy for increasing accuracy in settings when project circumstances are constantly changing.

Zitieren

APA

Ghosh, D., & Pandey, A. K. (2026). ADRL-CostNet: A Deep Reinforcement Learning Framework for Continuous and Self-Adaptive Software Cost Estimation in Agile Environments. International Journal of Drug Delivery Technology, 16(35s). https://doi.org/10.25258/ijddt.16.35s.88

Vancouver

Ghosh D, Pandey AK. ADRL-CostNet: A Deep Reinforcement Learning Framework for Continuous and Self-Adaptive Software Cost Estimation in Agile Environments. International Journal of Drug Delivery Technology. 2026;16(35s). doi:10.25258/ijddt.16.35s.88

BibTeX
@article{Ghosh2026,
  author = {Ghosh, Debanjan and Pandey, Arvind Kumar},
  title = {ADRL-CostNet: A Deep Reinforcement Learning Framework for Continuous and Self-Adaptive Software Cost Estimation in Agile Environments},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {35s},
  doi = {10.25258/ijddt.16.35s.88}
}
Kostenlos starten →