Internet of Things Based Weekly Crop Pest Prediction by Using Deep Neural Network

Abstract

Internet of Things (IoT) assisted application in agriculture shows tremendous success to improve productivity in agriculture. Agriculture is grappling with issues such as depleted soil fertility, climate-related hazards like intensified pest attacks and diseases. Accurate forecasting of pest outbreaks can play a vital role in improving agricultural yield. Utilizing IoT technology for environmental monitoring in crop fields to forecast pest attacks. The important parameters for pest predictions are temperature, humidity, rainfall, wind speed and sunshine duration. Directly sensed environmental conditions are utilized as input to a deep learning model, which makes binary decisions about the presence of pest populations based on the prevailing environmental conditions. The accuracy and precision of the deep learning model in making predictions are assessed through evaluation with test data. Five-year data 2028 to 2022 have been used for making prediction. The model of pest prediction generates weekly predictions. The overall accuracy of the weekly predictions is 94% and high F-measure, Precision, Recall, Cohens kappa, and ROC AUC for making to optimize the prediction. The accuracy of the pest prediction improves gradually with time. Weekly predictions are generated from the means of all environmental conditions from the last seven days. The weekly predictions are important for the short-term measures against pest attacks.