Predicting soil moisture based on the color of the leaves using data mining and machine learning techniques

Abstract

Abstract This research article’s aim is by using data mining and finding a suitable machine learning algorithm (MLA) to predict soil moisture, therefore the need for watering. Prediction is based on a training data set (including color RGB values taken from the leaves and values for soil moisture and soil temperature). A classifier is trained first, on its base a model is created and stored. Finally, with a different test data set, the efficiency of the selected model is checked. The object of study is the color of leaves of indeterminate greenhouse tomato plants of the Panekra variety. According to preliminary assumptions, the most informative about the need for watering are the young leaves (on top of the plant). Among the wide variety of data mining tools, we chose Weka Workbench. The last task of this study is to compare received with the methods of machine learning model and the model obtained in a previous study. For greater completeness of this research, the training of the classifier has been performed both with the whole training data set and with smaller data sets filtered by certain criteria (young/old leaves, etc.). The ultimate goal is water-saving and optimizing watering and water using. The resulting model is efficient and predicts soil moisture based on the color of the young leaves with 0-5% error, and by 8-12%, based on the color of the old ones, before watering, taking into account the influence of soil temperature into the model.