AI model predicts heart attack deaths with 98% accuracy using biomarkers

A new research paper from Sagnik Ghosh presents an automated AI model designed to predict lethal outcomes from acute myocardial infarction (MI) – commonly known as heart attacks. Cardiovascular disease remains a leading global cause of death, with MI claiming millions of lives annually. The model combines a deep artificial neural network with an ensemble of machine learning methods including Logistic Regression, Random Forest, LightGBM, and Bagging SVM. To address real-world clinical data challenges, it employs SVMSMOTE, ADASYN, and class-weighted techniques to handle imbalanced datasets, and uses wrapper and embedded feature selection to identify the most critical biomarkers linked to complications.

The system processes patient data through a pipeline that fills missing values, scales features, and applies advanced selection methods before classification. Early results suggest the ensemble plus ANN approach significantly improves prediction accuracy, precision, and recall compared to traditional clinical assessment. The goal is to provide doctors with a fast, consistent, and affordable tool for early thrombolytic treatment decisions, potentially reducing the 5-10% first-year mortality rate among MI survivors. While still in thesis stage, this work demonstrates how modern AI can transform critical care diagnostics by automating pattern recognition from complex medical data.