Machine Learning‐Based Air Temperature Prediction for Homa Bay County in Kenya: Implications for Weather‐Sensitive Sector Resilience

ABSTRACT This paper explores the challenges in temperature prediction, particularly in Homa Bay County, Kenya, a region with complex climatic conditions. Although traditional weather prediction models, such as numerical weather prediction, have limitations in regions with diverse topography, machine learning (ML), a data‐driven modeling approach, offers promising solutions for localized forecasting. Two‐year (2019–2021) meteorological data from Homa Bay County in Kenya were used to develop linear regression, decision tree, artificial neural network, random forest, and extreme gradient boosting ML models. The validation and test datasets were used to check the performance of the models. A systematic hyperparameter tuning was applied to improve predictive performance and generalization. A comparative evaluation of the five ML models was done. Random forest recorded the best performance with an R 2 , mean squared error, and mean bias error of approximately 0.91, 1.33, and 0.0031, respectively. The high influence of relative humidity and solar irradiance underscores the importance of atmospheric moisture and radiative processes in regional climate modeling. This study shows the potential of machine learning techniques in localized temperature prediction. However, the study is geographically and temporally bounded, which may limit the wider applicability of the results across regions with distinct climatic dynamics and environmental conditions. The findings in this study can support climate‐informed agricultural and water resource planning.