Comparison of machine learning models in outdoor temperature sensing by commercial microwave link

Abstract

The main objective of this work is to focus on outdoor temperature prediction using machine learning based on parameters from commercial microwave links. This information can be used to refine the weather information at a given link location. Three machine learning models (random forest, linear regression, and lasso) are used for prediction using a combination of two datasets (ERA5 weather dataset and CML monitoring database dataset). The results were evaluated based on two evaluation metrics (R^2 and mean absolute error (MAE)). In this work, the ERA5 outdoor temperature was found to be correlated with the temperature of the microwave link unit, and results were obtained with an accuracy of 0.87144 based on the MAE metric. Thus, the results can fairly well predict actual outdoor temperatures in the microwave link area based on the microwave link unit temperature.