Machine learning algorithms to assess the thermal behavior of a Moroccan agriculture greenhouse

Abstract The objective of this paper is to assess the potential of machine learning algorithms in predicting the indoor air temperature in a greenhouse using the outdoor data. A dataset gathering the main weather data and the indoor air temperature of a greenhouse located in Agadir, Morocco was used for this purpose. Machine learning models including support vector machine based-regression, ensemble trees and Gaussian process regression are compared against multiple linear regression models. This comparison was carried out on the basis of a 5-fold cross validation framework and across unseen data. The results show that all predictive models are capable of describing the indoor air temperature of the greenhouse and perform well (R2 > 0.9), with only 10% fraction of the dataset as training data. The Gaussian process regression outperforms all models, with R2 = 0.94 in the 5-fold cross validation test. However, the computational time related to the training of Gaussian process regression model is slightly higher than other machine learning models.