nexusbert/heart-disease-cnn

Heart Disease 1D CNN

A PyTorch 1D Convolutional Neural Network for binary heart disease classification from seven selected tabular features.

Model Description

This model was developed as part of a comparative machine learning study for heart disease classification.

The model uses a 1D convolutional architecture over seven preprocessed features.

Task

Binary classification:

• 0 — No Heart Disease
• 1 — Heart Disease

Input Features

The model expects seven preprocessed features in exactly this order:

Index	Feature
0	Age
1	Resting Blood Pressure
2	Cholesterol
3	Fasting Blood Sugar
4	Oldpeak
5	Exercise Angina: No
6	Exercise Angina: Yes

The numerical features must use the same preprocessing/scaling procedure used during training.

Architecture

The CNN contains:

1. Conv1D — 32 filters
2. Batch Normalization
3. ReLU
4. Conv1D — 64 filters
5. Batch Normalization
6. ReLU
7. Dropout
8. Flatten
9. Dense layer — 64 units
10. ReLU
11. Dropout
12. Binary output layer

Input shape:

(batch_size, 7)

and internally reshapes them to:

(batch_size, 1, 7)

for Conv1D processing.

Training Configuration

Parameter	Value
Optimizer	AdamW
Learning Rate	0.001
Weight Decay	1e-4
Batch Size	128
Early Stopping Patience	7
Loss	Binary Cross Entropy with Logits
Random Seed	42

Dataset Split

Split	Samples
Training	38,427
Validation	8,235
Test	8,235

The datasets were kept independent during model development and evaluation.

Evaluation

The CNN should be evaluated using:

• Accuracy
• Precision
• Recall
• F1 Score
• ROC-AUC
• PR-AUC
• Sensitivity
• Specificity

Final CNN test metrics should be added to this model card after the final CNN experiment.

Transformers Compatibility

This repository contains a custom Transformers configuration and model implementation.

The model can be loaded using Hugging Face Transformers with remote-code support.

Install dependencies:

pip install -r requirements.txt

Load the model:

import torch
from transformers import AutoModel


model_id = "nexusbert/heart-disease-cnn"


model = AutoModel.from_pretrained(
    model_id,
    trust_remote_code=True
)


model.eval()

Inference Example

import torch
from transformers import AutoModel


model_id = "nexusbert/heart-disease-cnn"


model = AutoModel.from_pretrained(
    model_id,
    trust_remote_code=True
)


model.eval()


features = torch.tensor(
    [[
        0.0,
        0.0,
        0.0,
        0.0,
        0.0,
        0.0,
        1.0
    ]],
    dtype=torch.float32
)


with torch.no_grad():
    outputs = model(
        input_features=features
    )


probability = torch.sigmoid(
    outputs["logits"]
).item()


prediction = int(probability >= 0.5)


label = (
    "Heart Disease"
    if prediction == 1
    else "No Heart Disease"
)


print("Prediction:", prediction)
print("Label:", label)
print("Probability:", probability)

Important Preprocessing Requirement

The model expects the seven features in their processed/scaled representation.

Raw clinical measurements should not be passed directly into the model unless they have first been transformed using the same preprocessing pipeline used during training.

For example, values such as:

• Age = 56
• Resting Blood Pressure = 145
• Cholesterol = 370

cannot automatically be assumed to be valid model inputs.

The exact preprocessing procedure must be reproduced before inference.

Intended Use

This model is intended for:

• Machine learning research
• Educational demonstrations
• Tabular classification experiments
• Benchmarking neural-network approaches
• Research into healthcare machine learning

Limitations

This model is a research and educational machine learning system.

It is not a medical diagnostic device and must not be used to diagnose, treat, or make clinical decisions about an individual.

Performance on the development dataset does not establish clinical validity or guarantee generalization to other populations or real-world clinical environments.

Ethical Considerations

Healthcare prediction systems can have significant consequences when incorrectly interpreted or deployed.

Any real-world clinical application would require appropriate external validation, clinical review, safety evaluation, monitoring, and regulatory assessment.

Repository Contents

• heart-disease-cnn/
  • config.json
  • pytorch_model.bin
  • modeling_heart_disease_cnn.py
  • inference.py
  • requirements.txt
  • README.md

Model Card Authors

Developed as part of a heart disease machine learning research project.

License