Alsanoosy et al. (2025) Predicting plant stress using SAM-L: novel self-adaptive-meta learner with XAI based on soil moisture and chlorophyll analysis

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-11-27
• Authors: Tawfeeq Alsanoosy, Javaid Ahmad Malik
• DOI: 10.1038/s41598-025-26184-w

Research Groups

• Department of Computer Science, College of Computer Science and Engineering, Taibah University, Madinah, Saudi Arabia
• Energy, Industry, and Advanced Technologies Research Center, Taibah University, Madinah, Saudi Arabia
• National College of Business Administration and Economics, Lahore (Main Campus), Lahore, Pakistan

Short Summary

This study proposed a novel framework integrating Sparse Additive Models with Learning (SAM-L) and Explainable Artificial Intelligence (XAI) to predict plant stress using soil moisture and chlorophyll content. The framework achieved an overall accuracy of 89.2% on a multi-class classification task, providing adaptive and interpretable stress predictions for precision agriculture.

Objective

• To develop a novel plant stress prediction framework that integrates Sparse Additive Models with Learning (SAM-L) and Explainable Artificial Intelligence (XAI) techniques to enhance prediction accuracy in smart farming, adaptively predicting plant stress based on soil moisture and chlorophyll content.

Study Configuration

• Spatial Scale: General agricultural environments, designed to generalize across diverse farms, crops, and weather patterns. The model was tested on a publicly available simulated dataset.
• Temporal Scale: Real-time and sequential data processing, with variables like soil moisture and chlorophyll content being time-dependent. The dataset included timestamps, and temporal variations were analyzed over periods like October to November 2024.

Methodology and Data

• Models used: Sparse Additive Models with Learning (SAM-L), Long Short-Term Memory (LSTM) network (three-layer, 256 neurons per layer), Explainable Artificial Intelligence (XAI) techniques including Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP). The model used ReLU activation for hidden layers and softmax for the output layer, optimized with Adam (learning rate 0.001) for 50 epochs with a batch size of 32.
• Data sources: A publicly available dataset titled "Real-Time Plant Health Insights: Simulated Biosensor Data for AI-Driven Monitoring," sourced from Kaggle (https://www.kaggle.com/datasets/ziya07/plant-health-data). The dataset includes attributes such as Timestamp, PlantID, SoilMoisture, AmbientTemperature, SoilTemperature, Humidity, LightIntensity, SoilpH, NitrogenLevel, PhosphorusLevel, PotassiumLevel, ChlorophyllContent, ElectrochemicalSignal, and PlantHealth_Status.

Main Results

• The proposed framework achieved an overall accuracy of 89.17% on the multi-class classification task (healthy, moderate stress, high stress).
• It demonstrated strong performance with a macro F1-score of 0.88, a macro recall of 0.88, and a macro precision of 88.39%.
• SHAP analysis revealed chlorophyll content as the most significant feature for predicting healthy plant status (mean(|SHAP value|) ≈ 0.22), followed by soil moisture (mean(|SHAP value|) ≈ 0.13).
• The learning curve showed that the model's performance improved with larger datasets, exhibiting reduced overfitting and enhanced generalization capability.
• Soil moisture and chlorophyll content were identified as reliable indicators of plant health, effectively predicting stress conditions.

Contributions

• Proposed a novel framework integrating SAM-L and XAI to provide an adaptive, accurate, and transparent solution for plant stress prediction in precision agriculture.
• Enhanced prediction accuracy and promoted sustainable farming practices by reducing water wastage and improving crop resilience through real-time, interpretable insights.
• Addressed key limitations of traditional plant stress prediction models, such as lack of scalability, adaptability to dynamic conditions, and interpretability.
• Incorporated XAI techniques (LIME and SHAP) to ensure transparent decision-making, enabling farmers and stakeholders to understand the rationale behind irrigation recommendations and other agricultural decisions.
• Utilized a self-adaptive learning framework (SAM-L) based on LSTM architectures, allowing the model to continuously adjust its parameters in response to dynamic environmental conditions without requiring a complete retraining cycle.

Funding

This research did not receive funding.

Citation

@article{Alsanoosy2025Predicting,
  author = {Alsanoosy, Tawfeeq and Malik, Javaid Ahmad},
  title = {Predicting plant stress using SAM-L: novel self-adaptive-meta learner with XAI based on soil moisture and chlorophyll analysis},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-26184-w},
  url = {https://doi.org/10.1038/s41598-025-26184-w}
}