lucas-lfp / heart-shaped-box__machine-learning-svm-classifier Goto Github PK

View Code? Open in Web Editor NEW

Developpement of a machine learning model (SVM classifier) for cardiovascular disease prediction. Deployed on a streamlit app.

Python 90.17% CSS 3.40% HTML 5.85% Dockerfile 0.57%

heart-shaped-box__machine-learning-svm-classifier's Introduction

Heart Shaped Box - SVM Classifier for Cardiovascular Diseases Detection

Overview

This project focuses on the development of a Support Vector Machine (SVM) classifier for the detection of cardiovascular diseases. The SVM model achieved an accuracy of 73.6% and is deployed on a Streamlit app, allowing users to test it with their own data. Model was deployed on a Streamlit app, with two main functionnalitites:

User can browse through a collection of dataviz to gain relevant insight on adult population characteristics, regarding cardiovascular diseases.
User can test the model with their own data to gain insight on their cardiovascular status.

--> Visit the streamlit app.

Objectives

To understand the relationships between clinical features and, particularly, their connection to the target feature through relevant visualisations.
To find a suitable machine learning model for this application
To evaluate the impact of hyperparameters on the model's performances
To put the model into production so it can treat user's input

Skills Used in This Project

Project was developped in Python, using HTML/CSS for formatting.
Libraries used in this project: Pandas, NumPy, SciPy, Matplotlib, Seaborn, Scikit-learn, joblib, streamlit.

Method

Development of this application was made following a four-steps plan, as described below.

Data Analysis
Model Selection - Several models were evaluated on this dataset: Logistic Regression, Random Forest, Support Vector Machines, K-Nearest Neighbors and AdaBoost. Support Vector Machines was selected for this project, as it presented the best accuracy (73.2%), a good rate of correct predictions for Class 0 (77.6%) and an acceptable rate for Class 1 (68.8%).
Hyperparameter Tuning - SVM model was tuned to improve its performances, several combinations of kernel, C and gamma were tested. RBF kernel, with C = 0.1 and gamma = 10 led to good and balanced performances, and was the best combination for a high recall without a significant loss in accuracy.
Deployment