Machine Learning Model for Heart Attack Risk Prediction

[COVER]Structured input for the risk-prediction prototype

[01]

Overview

This risk-prediction prototype compares four binary-classification algorithms on 158,355 observations with 21 predictors.

A selected Random Forest model is packaged as a Flask inference service that returns a class, probability, supporting factors, and global feature importance.

[02]

My Contribution

Prepared the dataset by encoding five categorical fields and applying MinMaxScaler.
Created a stratified 80:20 split and applied SMOTE to the training partition only.
Compared Random Forest, linear SVM, KNN, and Logistic Regression with RandomizedSearchCV focused on F1.
Serialized the selected artifacts with Joblib and deployed the Flask service through Docker on Ubuntu.

[03]

Key Technical Notes

The original data contains 158,355 rows, 22 columns, and no missing values.
The stratified split produced 126,684 training rows and 31,671 test rows before training-only SMOTE balancing.
Random Forest reached 71.93% accuracy, 64.12% precision, 68.15% recall, 0.6607 F1, and 0.8015 ROC-AUC.
Logistic Regression produced the highest compared F1 at 0.6618, while KNN produced the highest recall at 70.40%; Random Forest was selected for its leading accuracy and ROC-AUC.

[04]

Visual Evidence

Risk-prediction prototype input evidence — FIG.01Structured patient clinical input submitted to the inference workflow.

Risk-prediction prototype output evidence — FIG.02Inference response showing risk classification and probability output.

[05]

Tech Stack

Python
Scikit-learn
Pandas
SMOTE
Joblib
Flask
Docker

[06]

Status & Link

Prototype with deployed inference service

Source repository or production deployment is private.

[01]

Overview

This risk-prediction prototype compares four binary-classification algorithms on 158,355 observations with 21 predictors.

A selected Random Forest model is packaged as a Flask inference service that returns a class, probability, supporting factors, and global feature importance.

[02]

My Contribution

Prepared the dataset by encoding five categorical fields and applying MinMaxScaler.
Created a stratified 80:20 split and applied SMOTE to the training partition only.
Compared Random Forest, linear SVM, KNN, and Logistic Regression with RandomizedSearchCV focused on F1.
Serialized the selected artifacts with Joblib and deployed the Flask service through Docker on Ubuntu.

[03]

Key Technical Notes

The original data contains 158,355 rows, 22 columns, and no missing values.
The stratified split produced 126,684 training rows and 31,671 test rows before training-only SMOTE balancing.
Random Forest reached 71.93% accuracy, 64.12% precision, 68.15% recall, 0.6607 F1, and 0.8015 ROC-AUC.
Logistic Regression produced the highest compared F1 at 0.6618, while KNN produced the highest recall at 70.40%; Random Forest was selected for its leading accuracy and ROC-AUC.

[04]

Visual Evidence

[05]

Tech Stack

Python
Scikit-learn
Pandas
SMOTE
Joblib
Flask
Docker

[06]

Status & Link

Prototype with deployed inference service

Source repository or production deployment is private.