Constructing and Optimising RNN Models to Predict Fruit Rot Disease incidence in Areca Nut Crop Based on Weather Parameters

• Published in: IEEE access 2023-01, Vol.11, p.1-1
• Main Authors: Krishna, Rajashree, Prema, K V
• Format: Article
• Language: English
• Subjects: Agricultural products; Algorithms; Areca nut; Crops; Deep learning; Disease prediction; Diseases; Fruit Rot disease; Fungicides; Gated recurrent unit; Genetic algorithm; Genetic algorithms; Mathematical models; Meteorological data; Meteorology; Neural networks; Optimisation; Optimization; Optimization methods; Prediction algorithms; Predictive models; Rainfall; Root-mean-square errors; Weather; Weather forecasting
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3311477

A farmer faces several challenges associated with fruit rot disease in the areca nut crop. Weather factors, including rainfall and temperature, largely influence the disease severity and spread of infection in crops. Significant growth has been achieved using RNN models for time series forecasting in the past few decades, but these models are not applied to fruit rot disease prediction. This study introduces Vanilla GRU, Stacked GRU, Bidirectional GRU, and Bidirectional LSTM models for the inaugural prediction of fruit rot disease scores using past weather data. The current investigation also involves the mitigation and comparison of forecast inaccuracies by utilising weight optimisation algorithms like Adam, Adagrad, RMSprop, and Genetic algorithms. Meteorological and disease score data are gathered from the Agricultural Research Station in Brahmavar, Karnataka, and CPCRI Kasargod, Kerala. These datasets are combined using a rule-based algorithm to train and evaluate the proposed model. Empirical outcomes reveal that the vanilla GRU model, when fine-tuned with the Adam algorithm, exhibits a diminished Mean Squared Error (MSE) value of 0.0009, an exceptionally minimal Mean Absolute Error (MAE) value of 0.02, and an elevated R-squared (R2) score of 0.99. Similarly, the Bidirectional LSTM model, optimised through RMSprop, yields an impressively low Root Mean Squared Error (RMSE) value of 0.033. Generally, the optimised Deep Learning (DL) models consistently demonstrate enhanced predictive precision compared to alternative models. In conclusion, the anticipation of areca nut disease, as facilitated by this study, stands to aid farmers in curtailing unnecessary fungicide application and achieving more favourable yields.