Predicting Heart Stroke Risk Using Neural Networks: A Data Science Approach

By Felipe Leite In the ever-evolving field of data science, predictive modeling has become a key tool in addressing health-related concerns. Recently, I undertook a project focused on predicting heart stroke risk using a neural network (NN) model. In this blog post, I’ll walk you through the process, key findings, and areas for future improvement.

Project Overview

The goal was to develop a predictive model that could assess an individual's likelihood of experiencing a heart stroke based on factors such as age, blood pressure, cholesterol levels, heart rate, and blood flow. Leveraging a dataset that included various medical indicators, I trained a neural network to make predictions and evaluate its performance.

Key Findings

1. Age as a Key Predictor: As shown in the results, age is a significant factor in heart stroke risk. The probability of experiencing a heart attack increases markedly after the age of 65, as supported by research from the National Institutes of Health.

2. Blood Pressure and Heart Stroke Risk: Higher resting blood pressure levels correlate with increased heart stroke risk. The Heart and Stroke Foundation of Canada (2023) outlines that individuals with blood pressure readings of 135 mm Hg or higher are at high risk, which aligns with our model’s predictions.

3. Cholesterol and Heart Stroke: Cholesterol’s impact on heart stroke risk showed a non-linear pattern. The University of Pennsylvania suggests a higher risk at cholesterol levels above 240 mg/dl, which our model captured effectively.

4. Max Heart Rate: Surprisingly, the analysis revealed that higher maximum heart rates were associated with a lower probability of heart stroke, likely due to increased oxygen circulation in the body.

5. Blood Flow Analysis: Poor blood circulation, indicated by peaks in inadequate blood supply, was a strong predictor of heart stroke, especially in individuals with higher cholesterol and artery blockages.
   

Model Performance

The NN model achieved an accuracy of 82%, with balanced precision and recall scores across different classes. However, predicting heart strokes presents challenges due to the subtlety of symptoms. The model’s recall for high-risk cases was lower than desired, indicating room for improvement.

Challenges and Future Improvements

1. Data Limitations: The dataset’s size limited the model's performance in detecting nuanced patterns. Future work will focus on incorporating more diverse datasets, including larger populations and additional health indicators (e.g., LDL and HDL cholesterol levels).

2. Model Optimization: I aim to enhance the model by experimenting with different architectures and hyperparameter tuning. Techniques such as ensemble modeling and advanced feature engineering could potentially improve precision.
   

Conclusion

Predictive models like this one offer valuable insights for healthcare professionals, potentially serving as a decision-support tool in medical diagnostics. As AI technology advances, integrating such models into clinical practice could help mitigate risks and save lives. Have you used AI in healthcare predictions? Share your experiences in the comments! You can find the full paper on the link below:

Final Team Assignment - Heart Attack Analysis (1).docx

Download DOCX • 818KB