An efficient IoT-based crop damage prediction framework in smart agricultural systems.

Area of Science:

• Agricultural Technology
• Data Science
• Machine Learning

Background:

• Smart agriculture relies on real-time data for effective crop management.
• Missing data in sensor networks poses a significant challenge to predictive modeling.
• Existing systems often lack robustness in handling incomplete agricultural datasets.

Purpose of the Study:

• To develop an efficient IoT-based framework for predicting crop damage using machine learning.
• To integrate Internet of Things (IoT) sensor data with ensemble learning (EL) for enhanced crop health forecasting.
• To address missing data issues through advanced imputation techniques within a decision support system.

Main Methods:

• Utilized Internet of Things (IoT) sensors for real-time data collection.
• Applied machine learning (ML) and ensemble learning (EL) techniques, including XGBoost, CatBoost, and LightGBM (LGBM).
• Implemented data imputation strategies using K-Nearest Neighbors, linear regression, and ensemble-based imputers, optimized with Bayesian Optimization.

Main Results:

• XGBoost achieved the highest accuracy (89.56%) and sensitivity (88.1%) for crop damage prediction.
• The XGBoost model demonstrated strong data imputation capabilities with MSE of 0.0213 and R-squared of 0.99.
• Ensemble learning classifiers like CatBoost (90.50% accuracy) and LGBM (90.23% accuracy) also provided competitive predictive performance.

Conclusions:

• The developed IoT framework offers a low-cost, power-efficient, and scalable solution for crop damage prediction.
• Integrating real-time IoT data with optimized ensemble learning significantly enhances smart farming capabilities.
• The framework's robust data imputation effectively addresses missing data challenges in agricultural datasets, improving model reliability.