EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Prompt engineering for large language models in test case generation

Anatolii Husakovskyi

Information Technology and Computer Engineering · 20.04.2026 · (23) · 22-34

DOI: 10.31649/vitce/1.2026.22 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Information Technology and Computer Engineering, online erschienen am 20.04.2026. Verfasst von Anatolii Husakovskyi. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The purpose of this study was to perform a three-scenario comparative analysis of the accuracy of intelligent fault diagnosis in complex technical systems using ship power plants (SPPs) as a representative case. The research sought to determine which diagnostic configuration ensures the highest accuracy and robustness under varying operational conditions. Three methodological configurations were analysed: a baseline model based on Case-Based Reasoning (CBR); CBR enhanced with probabilistic analysis using Bayesian networks and Markov chains; and a comprehensive integration of CBR with probabilistic models and simulation modelling of cascading failures. Experiments were conducted under three typical operational scenarios – nominal mode, high-load mode, and limited diagnostic data – reflecting real maritime conditions. Standard classification accuracy metrics were employed, including Accuracy, Recall, and F1-score. The results showed that the basic CBR configuration achieves an average accuracy of 82-85% under nominal conditions but significantly loses efficiency when data are incomplete. Integration with probabilistic models improves metric stability, increasing accuracy to 88-90%. At the optimal configuration of method weights (CBR – αd = 0.6, probabilistic models – βd = 0.2, simulation modelling – γd = 0.2), the minimum diagnostic error of 6% was achieved, and diagnostic accuracy exceeded 93% even under noisy or incomplete data. Analysis of confusion matrices and error visualisations revealed that integrated configurations reduce the number of type II errors by approximately 35% compared to isolated approaches. Three-dimensional plots of accuracy dependence on component weights confirmed a stable maximum in the balanced-parameter zone and highlighted the significance of the simulation component under complex operational conditions. The obtained results allowed formulating practical recommendations for selecting diagnostic configurations: CBR + Bayesian Networks for stable modes and full integration for overload or data-limited scenarios. The proposed methodology is adaptable to other intelligent diagnostic systems operating under uncertainty, variable load, and incomplete information, including cyber-physical and industrial systems. It represents a universal and scalable framework for applied diagnostics requiring high accuracy, adaptability, and robustness

Zitieren

APA

Husakovskyi, A. (2026). Prompt engineering for large language models in test case generation. Information Technology and Computer Engineering, 22-34. https://doi.org/10.31649/vitce/1.2026.22

Vancouver

Husakovskyi A. Prompt engineering for large language models in test case generation. Information Technology and Computer Engineering. 2026:22-34. doi:10.31649/vitce/1.2026.22

BibTeX
@article{Husakovskyi2026,
  author = {Husakovskyi, Anatolii},
  title = {Prompt engineering for large language models in test case generation},
  journal = {Information Technology and Computer Engineering},
  year = {2026},
  number = {23},
  pages = {22--34},
  doi = {10.31649/vitce/1.2026.22}
}
Kostenlos starten →