Crop Disease Prediction Using LSTM Networks with Attention Mechanism and IoT-Based Environmental Time-Series Data

Abstract

Accurate crop disease prediction is essential for mitigating agricultural losses and optimizing resource allocation. This study introduces a novel framework that combines Long Short-Term Memory (LSTM) networks with an attention mechanism to improve the predictive accuracy and interpretability of disease risk models. Leveraging environmental time-series data collected through IoT-enabled sensors, including temperature, humidity, rainfall, and soil conditions, the LSTM network effectively captures long-term dependencies, while the attention mechanism dynamically prioritizes critical time steps and features. Hyperparameter optimization using Bayesian techniques further enhances the model's robustness and generalization. Experimental results demonstrate that the proposed LSTM+Attention model achieves superior performance, with the lowest RMSE (0.142) and MAE (0.109), compared to traditional machine learning methods such as Random Forest and Support Vector Machines, as well as deep learning models like GRU. Furthermore, the model remains reliable under varying environmental conditions, including extreme temperatures and humidity levels, highlighting its adaptability to real-world agricultural scenarios. The findings of this study provide a robust and scalable solution for real-time disease risk prediction, offering significant value for precision agriculture by supporting intelligent decision-making, reducing crop losses, and improving farming sustainability.