Forecasting Harvest Yield with IoT-enabled Sensor Data of Malaysia Weather Conditions using Multiple Linear Regressions

Abstract

Harvest yields are influenced by weather conditions due to climate change. Our literature studies showed that only a few studies examined tropical climates and harvest yield forecasts. Therefore, sixteen-month real-world tropical outdoor climate data from IoT-enabled sensors were investigated to build predictive harvest yield models. This preliminary study aims to discover if weather conditions, such as water temperature, acidity, radiation, air temperature, pH, light intensity, and air humidity, show a statistically significant impact in estimating the harvest yield. In the modelling process, Multiple Linear Regression modelling techniques were used to build and compare the models. The models’ performance was evaluated according to the adjusted $\mathrm{R}^{2}$ and Root Mean Square Error (RMSE) measurements. It is found that water temperature and electricity conductivity significantly influenced different plant varieties’ yield in weight, contributing to 13% (i.e., adjusted R-square) of the variance of the harvest yield forecast, indicating the magnitude of characteristics that can explain the harvest yield. Furthermore, the Regression models showed that electricity conductivity has a statistically negative impact, whereas water temperature positively affects the yields. Our future work will use more data and different machinelearning algorithms to improve the model’s predictive power.