Predictive models for Lettuce quality from Internet of Things-based hydroponic farm

摘要

As changes in the environment affect quality and quantity of crop, yield forecasting becomes very important for farmers. Thailand's economy depends on agriculture for a very long time, and lettuce is sold at high price. The authors had deployed smart hydroponic lettuce farms using the Internet of Things to collect environmental data and control the farm's operation in real time. The experiment generated large dataset which was used to create regression models using machine learning techniques in this study. Features used include environmental data such as the amount and intensity of light, humidity, temperature, together with weekly measurement of plant growth. Target variables are total fresh weight, nitrate content, number of leaf, and leaf area. RMSE was used as a measure for model selection. Models were created using several linear regression techniques, as well as newer techniques, such as, SVR, MLR, and ANN and the best ones were selected for each week prediction. Using only environmental data up to the 3rd week of planting, our predictive model performs 24.44% better than SVR when predicting total fresh weight, 13.93% better when predicting nitrate content, 0.47% better for number of leaves and 12.04% better for leaf area. When using additional plant growth data, our model is 13.09%, 5.52%, 19.47% and 5.06% better than SVR when predicting total fresh weight, nitrate content, number of leaf and leaf area, consecutively.