Machine Learning-Based Forecasting of Outlet Air Temperature in an Earth-to-Air Heat Exchanger

Abstract

Earth-to-air heat exchangers (EAHEs) exchange sensible heat between flowing air and the surrounding ground. Forecasting outlet air temperature over short horizons may support monitoring and future supervisory control studies. This study benchmarks four established regression models—Linear Regression, Decision Tree, Random Forest, and HistGradientBoosting—for one-, three-, and six-hour-ahead prediction using an hourly EAHE analysis dataset. A persistence forecast, which carries the latest measured outlet temperature forward, provides a reference. The 3,120 observations span 9 November 2023 to 17 March 2024. Data were ordered chronologically and divided into training and testing segments using an 80:20 split; median imputation was fitted within the training pipeline. Model performance was assessed using mean absolute error (MAE), root mean squared error (RMSE), and the coefficient of determination (R²). Random Forest produced the best one-hour result (MAE 0.0578 °C, RMSE 0.0815 °C, R² 0.9855) and the best three-hour result (MAE 0.1974 °C, RMSE 0.2755 °C, R² 0.8341). At six hours, Linear Regression was the strongest evaluated regressor (MAE 0.3750 °C, RMSE 0.5124 °C, R² 0.4257). Errors increased with forecast horizon, and the best model depended on lead time. The contribution is a transparent, dataset-specific comparison rather than a new learning algorithm. External validation and direct measurement of energy or comfort outcomes are required before practical benefits can be established.

Introduction

Machine Learning-Based Prediction of Earth-to-Air Heat Exchanger (EAHE) Outlet Temperature focuses on predicting the outlet air temperature of an Earth-to-Air Heat Exchanger using machine-learning techniques. EAHE systems use the natural temperature of the ground to heat or cool incoming air through underground pipes, helping improve indoor thermal comfort and reduce energy consumption.

The proposed study uses historical hourly weather and temperature data to predict future outlet temperatures at 1-hour, 3-hour, and 6-hour intervals. Four machine-learning regression algorithms are evaluated: Linear Regression, Decision Tree, Random Forest, and HistGradientBoosting. Their performance is also compared with a persistence model, which assumes that the latest measured outlet temperature remains unchanged.

The dataset contains 3,120 hourly observations collected between November 2023 and March 2024. It includes outlet temperature, inlet temperature, ground temperature, humidity, air pressure, and wind-related measurements. Solar radiation variables were excluded because approximately 83.04% of their values were missing.

The methodology involves data preprocessing, chronological splitting of training and testing data, model development, and performance evaluation using three metrics: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R²). The first 80% of the data is used for training, while the remaining 20% is used for testing.

The initial one-hour-ahead prediction results show that Random Forest achieved the best performance, with an MAE of 0.0578°C, RMSE of 0.0815°C, and R² of 0.9855. It also performed better than the persistence baseline, reducing MAE by 35.8% and RMSE by 20.8%.

Conclusion

This study compared Linear Regression, Decision Tree, Random Forest, and HistGradientBoosting for EAHE outlet-temperature forecasting at 1-, 3-, and 6-hour horizons, using a chronological 80:20 split and a persistence baseline. Random Forest provided the best one-hour and three-hour results, whereas Linear Regression was the strongest tested regressor at six hours. The increasing errors with lead time indicate that forecast difficulty depends on the prediction horizon. The contribution is a reproducible, dataset-specific benchmark of established models rather than a novel algorithm. The results should be interpreted within the limitations of a single analysis table, missing solar radiation data, and a single hold-out period. Future work should verify raw-data lineage, use rolling-origin and independent-site validation, improve feature coverage, and assess uncertainty. Any claims about energy use, thermal comfort, or operational benefit require separate measurement or validated system-level analysis.

References

[1] Bisoniya, T. S., Kumar, A., & Baredar, P. (2014). Study on calculation models of earth-air heat exchanger systems. Journal of Energy, 2014, Article 859286. https://doi.org/10.1155/2014/859286 [2] Ahmed, S. F., Liu, G., Mofijur, M., Azad, A. K., Hazrat, M. A., & Chu, Y.-M. (2021). Physical and hybrid modelling techniques for earth-air heat exchangers in reducing building energy consumption: Performance, applications, progress, and challenges. Solar Energy, 216, 274–294. https://doi.org/10.1016/j.solener.2021.01.022 [3] Hasan, M. I., Noori, S. W., & Shkarah, A. J. (2019). Parametric study on the performance of the earth-to-air heat exchanger for cooling and heating applications. Heat Transfer—Asian Research, 48(5), 1805–1829. https://doi.org/10.1002/htj.21458 [4] Ahmad, S. N., & Prakash, O. (2023). Three-dimensional simulation of earth air heat exchanger for cooling application and its validation using an experimental test rig. Heat and Mass Transfer, 59(7), 1277–1292. https://doi.org/10.1007/s00231-022-03334-8 [5] Shukla, A., Tiwari, G. N., & Sodha, M. S. (2006). Parametric and experimental study on thermal performance of an earth-air heat exchanger. International Journal of Energy Research, 30(6), 365–379. https://doi.org/10.1002/er.1154 [6] Alidrissi, Y., Wakil, M., Najib, M., Kaitouni, S. I., Mghazli, M. O., & Bakhouya, M. (2024). EAHE-D: A dataset for modeling and performance evaluation of earth to air heat exchangers. Data in Brief, 57, 111110. https://doi.org/10.1016/j.dib.2024.111110 [7] Kifouche, A., Kaddour, A., Lalmi, D., Chenini, N., Alkhafaji, M. A., Chambashi, G., Kaid, N., & Menni, Y. (2024). Enhancing real-time earth–air heat exchanger outlet temperature forecasting in arid climates using artificial neural network: A case study from Bechar, Algeria. International Journal of Low-Carbon Technologies, 19, 2493–2501. https://doi.org/10.1093/ijlct/ctae206 [8] Yue, Y., Yan, Z., Ni, P., Lei, F., & Yao, S. (2024). Machine learning-based multi-performance prediction and analysis of Earth-Air Heat Exchanger. Renewable Energy, 227, 120550. https://doi.org/10.1016/j.renene.2024.120550 [9] Yue, Y., Yan, Z., Ni, P., Lei, F., & Zhang, D. (2025). Meteorological effects on the thermal performance of earth–air heat exchanger: Insights from interpretable machine learning. Energy and Buildings. https://doi.org/10.1016/j.enbuild.2025.116304 [10] Qi, X., Yang, D., Guo, X., Chen, F., An, F., & Wei, H. (2024). Theoretical modelling and experimental evaluation of thermal performance of a combined earth-to-air heat exchanger and return air hybrid system. Renewable Energy. https://doi.org/10.1016/j.renene.2024.121418 [11] Ahmad, S. N., Prakash, O., & Saini, P. (2025). Performance investigation of earth air heat exchanger: A synergistic approach of 3D simulation and its experimental validation for heating mode. Environmental Progress & Sustainable Energy, 44(4), e14620. https://doi.org/10.1002/ep.14620 [12] EAHE-D dataset. (2024). EAHE-D: A dataset for modeling and performance evaluation of earth to air heat exchangers. Zenodo. https://doi.org/10.5281/zenodo.12820831

Copyright

Copyright © 2026 Gubbala Bindu Madhavi, Ravi Kumar Boddu, Lakshmi Tulasi Chavali, Suresh Kumar Mandru. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.