An Effective Disease Detection Analysis on Rice Leaves Using Hybrid MCSVM-DNN Predictor Architecture

Rice plant diseases significantly threaten global food security by affecting various growth stages. Timely and accurate prediction is vital for implementing control measures to mitigate yield losses. Traditional deep neural network (DNN) models, however, struggle to provide reliable predictions for rice plant diseases due to their complex and diverse symptoms. In this study, we introduce a novel Hybrid Multi-Class Support Vector Machine-Deep Neural Network (MCSVM-DNN) framework, which accurately classifies diseases based on symptoms observed at each rice plant stage. The hybrid MCSVM-DNN predictor architecture combines the strengths of MCSVM, which optimizes features after training with enough rice images at each stage, and DNN, which supports dynamic updating of extracted features. This combination results in a robust and efficient disease prediction system. We evaluate our model’s performance using a comprehensive dataset of rice leaf images from various disease types and growth stages, sourced from the freely available Kaggle online resource. Our hybrid MCSVM-DNN predictor demonstrates superior performance with higher recall, precision, F1-score, and accuracy values compared to existing DNN-based rice plant disease prediction models. Overall, the proposed architecture significantly improves disease prediction accuracy and reliability, contributing to global food security and sustainable agriculture.