Hadoop–Spark Framework for Machine Learning-Based Smart Irrigation Planning

Abstract

Up-to-date, given the expanding increase of the population and the development of human daily lifestyles, the expenditure of freshwater resources increments progressively. It appears that there is a need to optimize at least the consumption of fresh water in agriculture. For this reason, novel various irrigation technologies have been deployed in this context like drip irrigation, flood irrigation, and decision support systems to come up with the constraints of climate changes that decrease the water availability but it is still limited. Therefore, the majority of researchers are working until today on automating the irrigation systems. These smart systems rely mainly on the advances of information technologies like the internet of things, big data, and machine learning for aligning irrigations with climatic changes. Besides, integrating the predictive process helps in anticipating and adapting to the climatic constraints in agriculture, using meticulous soil and environment dependencies analysis based on features' prediction. In this paper, we enriched our proposed flexible online learning (OL) framework designed for promoting irrigation decisions based on soil characteristics analysis and prediction. We shed the light on a comparative study of four predictive methods, in particular, the auto-regressive moving average, the eXtreme Gradient Boosting, the random forest, and the deep artificial neural networks implemented inside the Hadoop/Spark environment to predict the humidity of the soil, relying on soil temperature and time in several depths. In the end, we discussed the precision of these models in various conditions.