Classification of toxic element accumulation in rice grains using optimized machine learning models: A comparative study

Md Imran Ullah Sarkar, Asadi Srinivasulu, Alvin Lal, Ravi Naidu, Mohammad Mahmudur Rahman,
Classification of toxic element accumulation in rice grains using optimized machine learning models: A comparative study,
Journal of Food Composition and Analysis,
Volume 148, Part 4,
2025,
108504,
ISSN 0889-1575,
https://doi.org/10.1016/j.jfca.2025.108504.
(https://www.sciencedirect.com/science/article/pii/S0889157525013201)
Abstract: The presence of essential and toxic elements in rice grains raises significant concerns regarding food safety and public health. This study conducts a comparative evaluation of five machine learning (ML) models - Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), XGBoost (Gradient Boosting), and Artificial Neural Networks (ANN) to predict the accumulation of toxic elements in rice. A dataset comprising 144 rice samples from various regions of Bangladesh was analyzed to examine the interactions between essential elements (Co, Cu, Fe, Mn, Mo, Se, and Zn) and toxic elements (As, Cd, Pb, Ni, and Cr). The experimental results indicate that LR (75.86 %) and ANN (72.41 %) exhibited the highest accuracy, surpassing other models. SVM (65.52 %) performed moderately well, whereas RF (55.17 %) and XGBoost (44.83 %) faced challenges related to overfitting and dataset constraints. Feature importance analysis using RF and XGBoost identified key essential elements contributing to toxic element accumulation. Principal Component Analysis (PCA) was utilized to visualize geographical variations in elemental distribution. The study underscores that while tree-based models like RF provide interpretability, and deep learning approaches such as ANN demonstrate superior predictive capabilities. These findings support the advancement of data-driven agricultural strategies, assisting policymakers and researchers in mitigating toxic element risks in rice and improving food safety. Future studies should focus on expanding datasets, leveraging advanced deep learning techniques, and incorporating additional environmental factors to enhance predictive modeling accuracy further.
Keywords: Rice grains; Machine learning techniques; Toxic element accumulation; Random Forest (RF); Support Vector Machine (SVM); Artificial Neural Networks (ANN); XGBoost (Gradient Boosting); Food safety; Feature importance analysis