Physics-Aware Ensemble Learning for Superior Crop Recommendation in Smart Agriculture

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Agriculture is crucial for national economies, necessitating advancements in crop recommendation systems for enhanced productivity and resource management.
• Traditional machine learning (ML) and ensemble learning (EL) models for precision farming face limitations with noisy or limited training data, leading to inaccurate predictions.
• Integrating physical laws into ML frameworks can ensure predictions are physically plausible, overcoming limitations of data-dependent models.

Purpose of the Study:

• To analyze and compare the performance of ML and EL models against a novel physics-informed ML model for agricultural applications.
• To introduce a stacking physics-informed ML model that incorporates crop-specific physical laws (optimal temperature and pH) during training.
• To evaluate the model's ability to simultaneously fit data and adhere to physical constraints.

Main Methods:

• Development of a stacking physics-informed ML model incorporating optimal crop temperature and pH as physical constraints.
• Training the physics-informed model with a dual objective: fitting training data and satisfying physical laws via a penalty term in the loss function.
• Comparative analysis of the proposed model against standard ML and optimized EL models.

Main Results:

• The proposed stacking physics-informed ML model achieved a highest accuracy of 99.50%.
• This performance significantly surpasses that of traditional ML and EL models, even those with optimization.
• The physics-informed approach ensures predictions are both accurate and physically feasible.

Conclusions:

• Physics-informed ML models offer a superior approach to crop recommendation and precision farming by integrating domain knowledge.
• The novel stacking physics-informed model demonstrates high accuracy and reliability, addressing limitations of purely data-driven methods.
• This methodology holds significant potential for advancing agricultural productivity and sustainable resource management.