Investigating the Synthetic Minority class Oversampling Technique (SMOTE) on an imbalanced cardiovascular disease (CVD) dataset

TL;DR

This study evaluates the impact of using SMOTE for data augmentation on an imbalanced cardiovascular dataset, highlighting its potential benefits and limitations in improving machine learning classification performance.

Contribution

The paper investigates the effectiveness of SMOTE in augmenting an imbalanced CVD dataset and compares classifier performance before and after augmentation.

Findings

SMOTE can improve classifier accuracy in some cases

Data augmentation with SMOTE is not universally effective

Careful analysis is needed before applying SMOTE

Abstract

In this work, we employ the Synthetic Minority Oversampling Technique (SMOTE) to generate instances of the minority class of an imbalanced Coronary Artery Disease dataset. We firstly analyze the public dataset Z -- Alizadeh Sani, a dataset used for non-invasive prediction of CAD. We perform feature selection to exclude attributes unrelated to Coronary Artery Disease risk. The generation of new samples is performed using SMOTE, a technique commonly employed in machine learning tasks. We design Artificial Neural Networks, Decision Trees, and Support Vector Machines to classify both the original dataset and the augmented. The results demonstrate that data augmentation may be beneficial in specific cases, but it is not a panacea, and its application in a specific dataset should be carefully examined.