EUVIMEDEUVIMED entdecken
Crossref

From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales

Cedric Beogo, Wepari Yaguibou, Eric Korsaga

Nuclear Science · 13.11.2026 · 11 (2) · 20-29

DOI: 10.11648/j.ns.20261102.11 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Nuclear Science, online erschienen am 13.11.2026. Verfasst von Cedric Beogo und 2 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Radon is the leading cause of lung cancer after cigarette smoking, and studies conducted in Burkina Faso have demonstrated that its concentration can reach relatively high levels inside residential dwellings due to the presence of radiological anomalies. This concentration varies significantly, making it highly complex to understand its behavior. In the present study, predictive models were trained to estimate radon concentrations based on environmental variables such as temperature, relative humidity, and atmospheric pressure. Continuous and simultaneous measurements were carried out inside a single dwelling in the village of Sissene using a Corentium Pro monitor from May 5 to September 6. An initial Random Forest model developed using hourly data yielded unsatisfactory predictive performance, with R 2 scores fluctuating between -0.8 and 0.2, specifically failing during the hourly transition periods at 06:00 and 16:00. Conversely, a second modeling approach using daily averaged data for both Random Forest and Gradient Boosting algorithms produced substantially better results. The highest predictive performances were achieved by the Random Forest model, with R 2 score and RMSE of 0.392 and 12.01 Bq∙m -3 respectively, compared to the Gradient Boosting model, which obtained values of 0.277 and 13.09 Bq∙m -3 respectively. Furthermore, the ablation study identified relative humidity as the most critical predictive variable for radon estimation, followed by temperature, whereas atmospheric pressure acted as a disruptive variable for the RF architecture. For the RF model, removing humidity and temperature caused the R 2 score to drop to 0.15 and 0.13 respectively, while removing pressure led to 0.08 increase. For the GB model, a massive impact was observed: the R 2 score dropped to 0.49, 0.30 and 0.04 upon the removal of humidity, temperature and pressure respectively.

Zitieren

APA

Beogo, C., Yaguibou, W., & Korsaga, E. (2026). From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales. Nuclear Science, 11(2), 20-29. https://doi.org/10.11648/j.ns.20261102.11

Vancouver

Beogo C, Yaguibou W, Korsaga E. From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales. Nuclear Science. 2026;11(2):20-29. doi:10.11648/j.ns.20261102.11

BibTeX
@article{Beogo2026,
  author = {Beogo, Cedric and Yaguibou, Wepari and Korsaga, Eric},
  title = {From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales},
  journal = {Nuclear Science},
  year = {2026},
  volume = {11},
  number = {2},
  pages = {20--29},
  doi = {10.11648/j.ns.20261102.11}
}
Kostenlos starten →