Comparative Evaluation of Explainable Machine Learning Versus Linear Regression for Predicting County-Level Lung...

Area of Science:

• Public Health
• Epidemiology
• Machine Learning

Background:

• Lung cancer (LC) is a major cause of cancer mortality in the U.S.
• Accurate LC mortality prediction is vital for interventions and addressing health disparities.
• Explainable machine learning offers potential for improved prediction and insight over traditional models.

Purpose of the Study:

• To predict county-level lung cancer mortality rates across the U.S.
• To compare the performance of random forest (RF), gradient boosting regression (GBR), and linear regression (LR) models.
• To identify key factors influencing LC mortality and analyze geographic disparities.

Main Methods:

• Applied RF, GBR, and LR models to predict U.S. county-level LC mortality.
• Evaluated model performance using R-squared and root mean squared error (RMSE).
• Utilized Shapley Additive Explanations (SHAP) for variable importance and Getis-Ord (Gi*) for spatial hotspot analysis.

Main Results:

• The RF model achieved the highest predictive accuracy (R-squared: 41.9%, RMSE: 12.8).
• Key predictors included smoking rate, median home value, and Hispanic population percentage.
• Significant LC mortality hotspots were identified in the mid-eastern U.S. counties.

Conclusions:

• The RF model provides superior prediction of LC mortality rates.
• Smoking prevalence, housing values, and Hispanic population percentage are critical factors.
• Findings support targeted interventions and health disparity reduction strategies in high-risk areas.