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 R2 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 R2 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 R2 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 R2 score dropped to 0.49, 0.30 and 0.04 upon the removal of humidity, temperature and pressure respectively.
| Published in | Nuclear Science (Volume 11, Issue 2) |
| DOI | 10.11648/j.ns.20261102.11 |
| Page(s) | 20-29 |
| Creative Commons |
This is an Open Access article, distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium or format, provided the original work is properly cited. |
| Copyright |
Copyright © The Author(s), 2026. Published by Science Publishing Group |
Radon Prediction, Random Forest, Gradient Boosting, Atmospheric Parameters, Burkina Faso, Indoor Air Quality, Corentium Pro
Parameters | Random Forest | Gradient Boosting |
|---|---|---|
Number of trees/ Boosting stage | 1000 | 1000 |
Maximum trees depth | 12 | 5 |
Minimum samples leaf | 3 | 5 |
Random seed | 42 | 42 |
Model |
| RMSE |
|---|---|---|
Random Forest | 0.392 |
|
Gradient Boosting | 0.277 |
|
GB | Gradient Boosting |
RF | Random Forest |
MSE | Mean Squared Error |
RMSE | Root Mean Square Error |
| [1] | Beogo C. E., Cisse O. I., Zougmore F. (2022) Assessment of radiological hazards from soil samples in the Northeastern area of Burkina Faso. SN Applied Sciences, 4(3): 73. |
| [2] | Beogo C. E, Cisse O. I., Ansong M., Zougmore F. (2024) Gamma‐Ray Spectrometry of Soil Samples From a Pilot Site of Uranium Anomaly in the Sahel Region of Burkina Faso. Applied and Environmental Soil Science, 2024(1): 1-14. |
| [3] | Beogo C. E, Cisse O. I. and Zougmore F. (2024) Spatial distribution of Cesium-137 in the Northeastern part of Burkina Faso, in: 2024 IEEE Multi-Conference on Natural and Engineering Sciences for Sahel’s Sustainable Development (MNE3SD), 1-6. |
| [4] | Derra M., Bambara L. T., Kaboré K., Sawadogo Y. Z., Cisse O. I., Zougmoré F. (2024) Measurement of Radon Concentration and Estimation of Cancer Risk in Twenty-Four Model Houses in the Town of Koudougou. Open Journal of Applied Sciences, 14(1): 193-204. |
| [5] | Bambara L. T., Kaboré K., Derra M., Beogo C. E., Cisse O. I., Zougmore F. (2021) Assessment of Indoor Radon Concentration in Residential Buildings at Ouagadougou and Estimation of the Annual Effective Dose. Radiation Science and Technology, 7(1): 41-46. |
| [6] | Bambara L. T., Doumounia A., Kohio N., Ouedraogo S., Zougmore F. (2022) Radon Level in Burkina Faso Student Residence and Estimation of the Annual Effective Dose. Asian Jounal of Scientific Research 15: 31‑38. |
| [7] | Rouamba J. M. M., Bambara L. T., Doumounia A., Kohio N., Ouedraogo S., Zougmore F. (2023) Evaluation of Radon Concentration and the Health Risk in the Offices of the Institute of Science and Technology of the “Ecole Normale Superieure”, Burkina Faso. World Journal of Nuclear Science and Technology, 13: 41-54. |
| [8] | Bambara L. T., Derra M., Kabore K., Cisse O. I., Zougmore F. (2023) Indoor Radon Concentration in Some Schools in Kaya, Burkina Faso. Physics and Materials Chemistry, 9(1): 8‑13. |
| [9] | Elola W. A., Bambara T. L., Doumounia A., Kohio N., Ouedraogo S., Zougmore F. (2023) Assessment of radon concentrations inside residential buildings and estimation of the dose in the city of Kaya, Burkina Faso. Open Journal of Applied Sciences, 13: 1066-1078. |
| [10] | Beogo C. E., Yaguibou W. C., Bambara, L. T. (2026) Spatial and temporal variability of radon and meteorological parameters in the dwellings of Villy. Sci Rep 16, 27886. |
| [11] | Beogo C. E., Yaguibou W. C., Bambara, L. T. (2026). Spatial and temporal variability of radon concentration with atmospheric parameters inside dwellings of Sissene, Burkina Faso. International Journal of Physical Sciences, 21(2), 29-40. |
| [12] | Chai T., Draxler, R. R. (2014) Root mean square error (RMSE) or mean absolute error (MAE)? – Arguments against avoiding RMSE in the literature. Geoscientific Model Development, 7(3), 1247-1250. |
| [13] | Geron A. (2019) Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. 2nd Edition, O’Reilly Media. |
| [14] | Chicco D, Warrens MJ, Jurman G. The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Comput Sci. 2021 Jul 5; 7: e623. |
APA Style
Beogo, C. E., Yaguibou, W. C., 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
ACS Style
Beogo, C. E.; Yaguibou, W. C.; Korsaga, E. From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales. Nucl. Sci. 2026, 11(2), 20-29. doi: 10.11648/j.ns.20261102.11
AMA Style
Beogo CE, Yaguibou WC, Korsaga E. From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales. Nucl Sci. 2026;11(2):20-29. doi: 10.11648/j.ns.20261102.11
@article{10.11648/j.ns.20261102.11,
author = {Cedric Elisee Beogo and Wepari Charles Yaguibou and Eric Korsaga},
title = {From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales},
journal = {Nuclear Science},
volume = {11},
number = {2},
pages = {20-29},
doi = {10.11648/j.ns.20261102.11},
url = {https://doi.org/10.11648/j.ns.20261102.11},
eprint = {https://article.sciencepublishinggroup.com/pdf/10.11648.j.ns.20261102.11},
abstract = {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 R2 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 R2 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 R2 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 R2 score dropped to 0.49, 0.30 and 0.04 upon the removal of humidity, temperature and pressure respectively.},
year = {2026}
}
TY - JOUR T1 - From Hourly Noise to Daily Signal: Machine Learning Modelling of Indoor Radon Dynamics Across Diurnal and Daily Scales AU - Cedric Elisee Beogo AU - Wepari Charles Yaguibou AU - Eric Korsaga Y1 - 2026/11/13 PY - 2026 N1 - https://doi.org/10.11648/j.ns.20261102.11 DO - 10.11648/j.ns.20261102.11 T2 - Nuclear Science JF - Nuclear Science JO - Nuclear Science SP - 20 EP - 29 PB - Science Publishing Group SN - 2640-4346 UR - https://doi.org/10.11648/j.ns.20261102.11 AB - 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 R2 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 R2 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 R2 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 R2 score dropped to 0.49, 0.30 and 0.04 upon the removal of humidity, temperature and pressure respectively. VL - 11 IS - 2 ER -