Environment-compatible heavy metal risk prediction method created with multilevel ensemble learning.

Area of Science:

• Environmental Science
• Toxicology
• Data Science
• Public Health

Background:

• Traditional health risk assessment for heavy metal (HM) exposure suffers from delays and passivity.
• Accurate health risk prediction (HRP) is crucial for mitigating HM exposure hazards.
• Existing methods may not fully integrate environmental compatibility for robust HRP.

Purpose of the Study:

• To propose an innovative health risk prediction (HRP) method, MEL-HR, integrating multilevel ensemble learning (MEL) and environmental compatibility.
• To evaluate the predictive performance of MEL-HR for health risks associated with HM exposure using point and interval predictions.
• To identify key environmental factors influencing HRP beyond HM concentrations.

Main Methods:

• Developed a multilevel ensemble learning (MEL) model (MEL-HR) incorporating environmental compatibility.
• Conducted point and interval prediction experiments using 490 datasets covering 17 environmental factors.
• Analyzed feature importance and performed comparative experiments against HM-only models.

Main Results:

• Point prediction achieved R² values of 0.707 for HI and 0.619 for TCR.
• Interval prediction (P5, P50, P95) showed R² values up to 0.706 for HI and 0.620 for TCR.
• Environment compatibility significantly improved accuracy by 19.83-40.36% (point) and 22.06-40.01% (interval) compared to HM-only models.
• Key environmental factors included longitude, mining area coefficient, and soil organic matter, alongside HM factors.

Conclusions:

• The proposed MEL-HR method effectively predicts health risks from HM exposure by integrating environmental compatibility.
• Environmental factors play a significant role in HRP, enhancing model accuracy beyond HM data alone.
• This approach offers technical support for targeted and resilient prevention and control of health risks.