Cardiovascular Risk Prediction After Viral Infection: A Random Forest-Based Machine Learning Medical Diagnosis System

A research project by KIIT University in India uses the random forest algorithm to build a machine learning model for predicting cardiovascular disease risk in patients with a history of viral infection, achieving an accuracy rate of 90.6% and published in the Springer Nature International Conference Proceedings.

• Original Author/Maintainer: Suchismita Sarkar (@suchismittaa), Simran Kumari, Srishti Sawarna
• Institution: School of Computer Engineering, KIIT University, Bhubaneswar, India
• Source Platform: GitHub
• Original Title: Cardiovascular-Impact-of-Viral-Infection-Prediction
• Original Link: https://github.com/suchismittaa/Cardiovascular-Impact-of-Viral-Infection-Prediction
• Paper Publication: Springer Nature, 5th International Conference on Computing and Communication Networks (ICCCNet-2025), Manchester Metropolitan University, UK
• Release Date: June 2, 2026

Cardiovascular disease (CVD) is one of the leading causes of death globally. Traditional cardiovascular risk assessment tools mainly focus on known risk factors such as smoking, diabetes, and hypertension, but often overlook an important factor: the long-term impact of past viral infections on the cardiovascular system.

Recent medical studies have shown that certain viral infections (e.g., influenza, COVID-19, etc.) may still have a lasting impact on the cardiovascular system after the acute phase, increasing the risk of future cardiovascular events in patients. However, existing clinical diagnostic methods—blood tests, electrocardiograms, and serological tests—do not include a history of viral infection in the routine risk assessment system.

This diagnostic gap leads to many high-risk patients being detected only after severe complications occur, missing the optimal timing for early intervention.

This study aims to fill this diagnostic gap by building a machine learning-based prediction model that combines viral infection history with clinical indicators to achieve early identification of cardiovascular risk in patients with a history of viral infection.

Core Contributions:

• For the first time, viral infection history is included as a feature in a cardiovascular disease prediction model
• Comparative evaluation of five mainstream machine learning algorithms on this task
• Achieved a prediction accuracy of 90.6%, which is better than traditional methods
• Provided an interpretable risk assessment framework to facilitate clinical decision support

The study uses the heart.csv dataset, which contains patient records covering viral infection history and cardiovascular health indicators. The dataset has undergone strict data cleaning, normalization, and feature scaling processes.

The model input features include:

• Demographic features: Age
• Physiological indicators: Blood pressure, cholesterol level, heart rate
• Medical history information: Past viral infection history
• Other clinical indicators: Relevant laboratory test results

1. Data cleaning: Handle missing values and outliers
2. Feature normalization: Eliminate dimensional differences to ensure all features are on the same scale
3. Feature scaling: Adjust feature distribution using standardization methods
4. Training/test split: Reasonable data division to ensure the model's generalization ability

The research team conducted a comparative evaluation of five mainstream machine learning algorithms:

Algorithm	Accuracy	Recall	F1 Score
Random Forest ⭐	0.906	0.844	0.879
Logistic Regression	0.852	0.844	0.857
Naive Bayes	0.869	0.844	0.871
Support Vector Machine	0.869	0.844	0.871
Decision Tree	0.754	0.656	0.737