Predicting heavy metal concentration in crop grain using automated machine learning models

Area of Science:

• Agricultural Science
• Environmental Science
• Data Science

Background:

• Heavy metal (HM) pollution in crops is a growing concern due to industrialization and intensive agriculture.
• Understanding HM accumulation in crops is crucial for food safety and sustainable agricultural practices.

Purpose of the Study:

• To predict HM concentrations in crop grains using automated machine learning (AutoML) models.
• To identify key factors influencing HM accumulation in crops.
• To evaluate the performance of various machine learning models for HM prediction.

Main Methods:

• Utilized 791 datasets from 54 publications to train AutoML models.
• Input variables included soil properties, fertilizer characteristics, and plant types.
• Output variables were concentrations of chromium (Cr), cadmium (Cd), lead (Pb), arsenic (As), and mercury (Hg) in crop grains.
• Evaluated deep learning (DL), gradient boosting machine (GBM), and other models.

Main Results:

• Deep learning (DL) models excelled in predicting Cr, Pb, As, and Hg.
• Gradient boosting machine (GBM) achieved the highest accuracy for Cd prediction.
• Organic fertilizer application and plant type were identified as primary drivers of HM accumulation.
• Positive correlations found with organic fertilizer application, soil HM concentration, and sand content.
• Negative correlations observed with cation exchange capacity, pH, organic matter, and clay content.

Conclusions:

• DL and GBM models demonstrate superior performance in predicting crop grain HM concentrations.
• Strict control over HM inputs in organic fertilizers is essential to mitigate risks in agriculture.