EUVIMEDEUVIMED entdecken
Open Access · CC BY-SA 4.0

Feature-Level Fusion for Predictive Drug Inventory Management in Hospitals Using XGBoost

Muhammad Furqan Nazuli, Kridanto Surendro

INSERT : Information System and Emerging Technology Journal · 25.12.2025 · 6 (2) · 183-197

DOI: 10.23887/insert.v6i2.101557 ↗

Kurz gefasst

Fachartikel in der Zeitschrift INSERT : Information System and Emerging Technology Journal, online erschienen am 25.12.2025. Verfasst von Muhammad Furqan Nazuli und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-SA 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Effective drug inventory management in hospitals remains a significant challenge due to fluctuating patient needs and dynamic external factors. Conventional approaches that rely only on historical usage data are often inadequate in predicting sudden changes, leading to overstock or stock-out, with negative impacts on patient safety and operational efficiency.

Objective

This study seeks to develop a predictive model for hospital drug inventory using a hybrid feature input and feature-level fusion, integrating historical drug usage, categorical drug information, and daily weather data.

Methods

A quantitative study was conducted using drug consumption records from January to December 2024. Feature engineering included numerical features (volume, moving averages), categorical features (drug types), and external features (temperature, weather). The XGBoost algorithm served as the main machine learning model. Model performance was evaluated with Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R², with cross-validation and hyperparameter tuning to optimize results.

Results

The optimized XGBoost model achieved an MAE of 409.51, RMSE of 1178.46, and R² of 0.8960 on the test set, indicating that 89.6% of the variance in actual drug demand was explained. The model’s predictions closely followed real trends, even during demand surges.

Conclusion

The proposed hybrid feature input model with feature-level fusion significantly improves predictive accuracy compared to traditional methods and provides a robust tool for hospital drug inventory management. Incorporating external data enhances adaptability to real-world conditions. Further research should explore real-time disease surveillance integration and evaluate the model’s generalizability across multiple healthcare settings.

Zitieren

APA

Nazuli, M. F., & Surendro, K. (2025). Feature-Level Fusion for Predictive Drug Inventory Management in Hospitals Using XGBoost. INSERT : Information System and Emerging Technology Journal, 6(2), 183-197. https://doi.org/10.23887/insert.v6i2.101557

Vancouver

Nazuli MF, Surendro K. Feature-Level Fusion for Predictive Drug Inventory Management in Hospitals Using XGBoost. INSERT : Information System and Emerging Technology Journal. 2025;6(2):183-197. doi:10.23887/insert.v6i2.101557

BibTeX
@article{Nazuli2025,
  author = {Nazuli, Muhammad Furqan and Surendro, Kridanto},
  title = {Feature-Level Fusion for Predictive Drug Inventory Management in Hospitals Using XGBoost},
  journal = {INSERT : Information System and Emerging Technology Journal},
  year = {2025},
  volume = {6},
  number = {2},
  pages = {183--197},
  doi = {10.23887/insert.v6i2.101557}
}
Kostenlos starten →