Air Pollution and Cardiovascular Mortality: An ML-Based Early Prediction Framework using Random Forest

Abstract

Responsible for an estimated 8.3 million premature deaths annually, of which more than half involve cardiovascular pathways. Fine particulate matter (PM2.5) — the most cardiotoxic ambient pollutant — penetrates alveolar tissue, enters systemic circulation, triggers oxidative stress and pro-inflammatory cytokine cascades (IL-6, TNF-α, CRP), destabilises atherosclerotic plaques, and precipitates acute myocardial infarction and cerebrovascular accidents. Despite overwhelming epidemiological evidence, clinical decision-support tools still rarely incorporate real-time environmental exposure data, creating a systematic blind spot in risk stratification. This paper presents a comprehensive dual-stream Machine Learning (ML) framework that fuses real-time ambient air quality telemetry — sourced from CPCB monitoring stations and WHO guidelines — with patient-level clinical parameters drawn from Electronic Health Records (EHR), to predict cardiovascular mortality risk at the individual level. A Random Forest ensemble classifier was selected as the primary model following rigorous benchmarking against four baseline algorithms (Logistic Regression, SVM, Decision Tree, and Neural Network). Experimental evaluation on a composite dataset of 2,500 subjects achieved an accuracy of 88.3%, precision of 87.1%, recall of 86.4%, and F1-score of 86.7%, outperforming all baselines. SMOTE oversampling corrected high-pollution exposure (≥20 years) was associated with a 68% cumulative increase in CVD mortality risk above baseline. class imbalance, and 5-fold stratified cross-validation ensured robust generalisation. PM2.5 concentration (Gini importance 0.31), systolic blood pressure (0.22), and patient age (0.18) were identified as the three most critical predictive dimensions