A Study on the Working Methodology of a From-Scratch Gaussian Naïve Bayes Algorithm for Heart Disease Classification Using a 12-Lead ECG Image Dataset

Parth Kailashbhai Thacker; Maheshkumar D. Mulani

doi:10.36948/ijfmr.2026.v08i05.87553

A Study on the Working Methodology of a From-Scratch Gaussian Naïve Bayes Algorithm for Heart Disease Classification Using a 12-Lead ECG Image Dataset

Author(s)	Mr. Parth Kailashbhai Thacker, Dr. Maheshkumar D. Mulani
Country	India
Abstract	Cardiovascular disease remains one of the leading causes of mortality worldwide, and electrocardiogram (ECG) interpretation continues to depend heavily on specialist availability, particularly in low-resource settings. This study presents a fully from-scratch implementation of the Gaussian Naïve Bayes (GNB) classifier for heart disease screening, evaluated on a publicly available 12-lead ECG image dataset comprising 929 images distributed across four classes: myocardial infarction (MI), history of MI, abnormal heartbeat, and normal ECG (Khan et al., 2021). Rather than relying on pre-digitized signal files, the study designs and applies a complete image-processing pipeline: grid and calibration-pulse removal, single-lead trace isolation, R-peak detection, and extraction of four physiologically motivated beat-level features (RR interval, QRS width, R-amplitude, and morphological mean). A GNB classifier implemented from first principles (per-class mean, variance, and log-space Gaussian likelihoods coded explicitly, without scikit-learn's naive_bayes module) was trained and validated on an 80/20 beat-level split. The baseline model achieved 50.33% validation accuracy; ANOVA F-value feature selection combined with a Yeo–Johnson power transformation modestly improved this to 50.88%, which is approximately double the four-class chance baseline of 25%. The results indicate that the RR interval and morphological mean are the most class-discriminative features, while confusion occurs primarily between the two myocardial infarction-related classes and the abnormal heartbeat/normal classes. The findings highlight both the interpretability advantages and the practical accuracy ceiling of a simple probabilistic classifier for image-derived ECG screening, and motivate hybridization with more expressive models.
Keywords	Gaussian Naïve Bayes, ECG classification, heart disease, machine learning, image digitization, myocardial infarction
Field	Computer Applications
Published In	Volume 8, Issue 5, September-October 2026
Published On	2026-09-12
DOI	https://doi.org/10.36948/ijfmr.2026.v08i05.87553

View / Download PDF File

About IJFMR Fees & Payment Current Issue Publication Archive	Submit Research Paper Track Submission Status Publication Guidelines Publication Ethics Peer Review & Plagiarism	Join as a Reviewer Editors & Reviewers Reviewer Referral Program Get Reviewer Membership Certi.	Website/Journal Policies Usage Policy Content Policies Privacy Policy

Contact Us		+91-9687-828-838	editor@ijfmr.com

International Journal For Multidisciplinary Research

A Widely Indexed Open Access Peer Reviewed Multidisciplinary Bi-monthly Scholarly International Journal

A Study on the Working Methodology of a From-Scratch Gaussian Naïve Bayes Algorithm for Heart Disease Classification Using a 12-Lead ECG Image Dataset

Share this