Analytics and Decision-making Model Using Machine Learning for Internet of Things-based Greenhouse Precision Management in Agriculture

Abstract

As wired systems for smart farming are difficult to manage and install, wireless connectivity is currently taking their place. Smart farming with precision greenhouse technology is installed to improvise in managing the growth of agriculture and therefore observing different environments in precision agriculture. Numerous systems have been developed for control and remote monitoring of precision agriculture. But due to limited solutions, monitoring of greenhouse is not yet competent to deal with the agricultural growth on entirely control systems. For better farming growth control, smart farming with precision greenhouses must be applied, necessitating precision agriculture monitoring under various circumstances. Supervised machine learning techniques are used in intelligent agricultural systems to provide intelligent information farming systems with predictive data analysis of sensor parameters. Cloud layer, fog layer, edge layer, and sensor layer are four important parts of the proposed approach. The data needed for the sensor layer of the analytical model is collected using Internet of Things-based embedded system devices in two greenhouse facilities, with sensor parameters as inputs and corresponding actuators as outputs. Using classification and regression models, two distinct analytical models for intelligent and accurate farming were built. By modifying farming circumstances in accordance with plant requirements taken into account during experimentation, the primary goals of this analytics are to boost output and offer organic farming. A decision-making and analytics system was built at the fog layer using the support vector machine and artificial neural network, two supervised classification-based machine learning methods. MATLAB software’s statistics and machine learning tools were used to analyze and interpret the experimental outcomes. Accuracy, sensitivity, specificity, and F-score are used to examine the confusion matrix-based metrics used in the performance evaluation of the suggested system. Based on the results of the experiments, the suggested method also proved to be the best at providing actuators with predictions and control.