Comparison of Machine Learning Regression Models for the Prediction of Soil Moisturewith the use of Internet of Things Irrigation System Data

Abstract

Internet of Things (IoT) technology allows the control and managementof systems independent of humans. IoT-based agriculture applications have becomewidespread as a solution to the problems of food consumption and water scarcity inagriculture as the world population has increased gradually. Soil moisture is animportant factor for agriculture production and hydrological cycles and theprediction of soil moisture is required in developing agricultural practices. In thisstudy, an IoT-based irrigation system prototype is presented which consists ofEsp8266 Wifi module, humidity and temperature, soil moisture, rain and ultravioletsensors connected to the Arduino Uno board. Using the prototype system, data arecollected from the pilot area determined in half-hour periods for 55 days and savedthe cloud with ThingSpeak. The soil moisture value is estimated by applyingdifferent machine learning regression models such as multiple linear, polynomial,support vector, decision tree and random forest regression using the collected data.To examine the success of the algorithms, the obtained results are comparedaccording to the coefficient of determination and the mean square error criteria. Itis found that the random forest regression model has found to be superior to othermachine learning algorithms for soil moisture estimation.