Probabilistic cancer risk assessment from heavy metal exposure in iranian rice and pasta: a novel hybrid framework integrating INAA, ICP-AES, and ensemble machine learning

This study investigates cancer risk from heavy metal exposure in rice and pasta using experimental data and machine learning approaches, based on 19 experimental samples and 1,750 simulated exposure instances. Concentrations of toxic heavy metals were quantitatively measured in multiple rice varieties and pasta types using Instrumental Neutron Activation Analysis (INAA) and ICP-AES analytical techniques. The experimentally determined metal concentrations were integrated into the Excess Lifetime Cancer Risk (ELCR) framework, and Several machine learning models were developed for sensitivity analysis and feature prioritization within the ELCR framework. Rather than predicting an unknown outcome (as ELCR is mathematically deterministic), the models were designed to quantify the relative contribution of each exposure parameter to overall cancer risk under probabilistic uncertainty. Regression analysis identified exposure duration as the most influential risk factor (R² = 0.263, p < 0.001), followed by chromium bioavailability (R² = 0.125) and pasta consumption patterns. Ensemble methods provided robust ranking of feature importance, demonstrating how machine learning can complement deterministic risk models by enabling multi-dimensional sensitivity analysis and uncertainty decomposition. The calculated ELCR values ranged from 1.2 × 10⁻⁶ (acceptable) to 1.8 × 10⁻⁴ (unacceptable) depending on consumption scenarios Cancer risk estimates spanned from acceptable to unacceptable levels depending on consumption scenarios. The integration of experimental analytical chemistry with machine learning provides a robust methodology for dietary cancer risk assessment. This approach offers reliable data for food safety regulations and public health protection.