Machine learning-based heart disease diagnosis: A systematic literature review

Area of Science:

• Cardiology
• Computer Science
• Data Science

Background:

• Heart disease remains a major global health challenge, with early detection crucial for effective management.
• Machine learning (ML) shows promise in diagnosing heart disease using electrocardiogram (ECG) and patient data.
• Imbalanced datasets in heart disease prediction hinder the unbiased performance of traditional ML algorithms.

Purpose of the Study:

• To systematically review and uncover challenges in heart disease prediction using imbalanced data.
• To analyze existing data-level and algorithm-level solutions for imbalanced data in cardiology.
• To provide a comprehensive overview of the current literature on ML for heart disease diagnosis.

Main Methods:

• Conducted a meta-analysis of 451 relevant articles published between 2012 and November 2021.
• Performed a systematic literature review (SLR) focusing on 49 selected articles.
• Analyzed articles based on heart disease type, ML algorithms, applications, and proposed solutions.

Main Results:

• Current ML approaches face significant open problems when handling imbalanced data in heart disease prediction.
• Existing methods often neglect critical aspects like interpretability and explainability of ML models.
• The practical applicability and functionality of ML in heart disease diagnosis are hindered by data imbalance issues.

Conclusions:

• Addressing data imbalance is critical for improving the reliability and utility of ML in heart disease diagnosis.
• Future research should focus on developing unbiased ML algorithms and solutions that consider interpretability.
• Enhanced ML strategies are needed to overcome limitations and improve data-driven decision-making in cardiology.