Comparative Study of Data Mining Techniques on Heart Disease Prediction System: a case study for the

Alladoumbaye Ngueilbaye, Lei Lin, Hongzhi Wang, A Nowbar, J Howard, J Finegold, P Asaria, D Francis, A Rajkumar, G Reena, A Roslina, A Noraziah, C Dangare, S Apte, A Sen, S Patel, D Shukla, D Medhekar, M Bote, S Deshmukh, S Pattekari, A Parveen, S Palaniappan, R Awang, R Wu, W Peters, M Morgan, B Boser, I Guyon, V Vapnik, G Graksn, H Hakl, H Uuz, C BurgesInternational Journal of Science and Research (IJSR) 2016 · DOI

Abstract

Nowadays, the healthcare sector is one of the areas where huge data are daily generated. However, most of the generated data are not properly exploited. Important encapsulated data are currently in the data sets. Therefore, this study presents a comparative analysis of two popular machine learning algorithms—Naïve Bayes and Support Vector Machines (SVM)—to predict heart disease using clinical data specifically sourced from the Republic of Chad.

Introduction

Heart disease remains a leading cause of death globally, with the World Health Organization (WHO) estimating millions of deaths annually. In the Republic of Chad, heart disease rates represent a significant public health concern. While modern healthcare facilities generate vast amounts of patient data, this information often remains "encapsulated" or underutilized. The diagnostic process for heart disease is complex and multi-layered, often prone to human error or subjective interpretation based solely on a clinician's experience. This research addresses these challenges by developing an automated, data-driven Clinical Decision Support System (CDSS) to assist medical professionals in making more accurate predictions.

Methodology

The study focuses on a comparative analysis of two primary classification techniques implemented within a web-based framework:

  1. Naïve Bayes (NB): A probabilistic classifier based on Bayes' Theorem. It assumes that all clinical features (e.g., age, blood pressure, cholesterol) are independent of one another. Despite this simplifying assumption, it is highly effective for large, complex medical datasets.
  2. Support Vector Machines (SVM): A supervised learning model that identifies an "optimal hyperplane" to separate classes (healthy vs. heart disease). This study utilized a Radial Basis Function (RBF) kernel to handle non-linear relationships by mapping variables into higher-dimensional space.
System workflow using the Naïve Bayes classifier

The researchers implemented these models using a hybrid technology stack: ASP.NET (C#) for the front-end user interface and Python for the backend data analysis and algorithmic processing.

Dataset and Implementation

The study utilized a localized dataset consisting of 315 instances collected from two major hospitals in Chad: Clinique Providence and Hôpital de Reference Nationale. The dataset was balanced, containing 150 instances from healthy patients and 165 from patients with heart disease. Each instance included 14 critical clinical attributes, such as age, gender, chest pain type, resting blood pressure, serum cholesterol, and fasting blood sugar.

To make these findings actionable, the authors developed a user-friendly web interface. This allows doctors to input patient parameters directly into the system to receive a real-time probability-based diagnosis.

Web-based user interface for clinical data entry

Results and Discussion

The performance of the two models was evaluated based on overall classification accuracy:

  • Naïve Bayes: 91.42% accuracy
  • Support Vector Machine: 89.56% accuracy

One of the most significant insights is that the simpler Naïve Bayes model outperformed the more complex SVM. This suggests that in medical diagnostics, where features are often treated as distinct clinical indicators, the independence assumption of Naïve Bayes does not necessarily hinder performance. Furthermore, the study emphasizes the importance of localized data; by using regional data from Chad, the model accounts for specific demographic and environmental factors that standard global datasets might miss.

Conclusion and Future Work

The experimental results clearly favored Naïve Bayes for this specific dataset, leading to its selection as the primary engine for the developed clinical support tool. However, the authors acknowledge limitations, including the relatively small dataset size (315 records). Future work will focus on expanding the dataset, incorporating more clinical attributes, and exploring additional algorithms such as Neural Networks or Decision Trees to further refine predictive power. The ultimate goal remains bridging the gap between raw medical data and practical clinical decision-making.

Summary updated

Abstract

Nowadays, the healthcare sector is one of the areas where huge data are daily generated. However, most of the generated data are not properly exploited. Important encapsulated data are currently in the data sets. Therefore,
BibTeX
@article{ngueilbaye2016comparative,
  author = {Alladoumbaye Ngueilbaye and Lei Lin and Hongzhi Wang and A Nowbar and J Howard and J Finegold and P Asaria and D Francis and A Rajkumar and G Reena and A Roslina and A Noraziah and C Dangare and S Apte and A Sen and S Patel and D Shukla and D Medhekar and M Bote and S Deshmukh and S Pattekari and A Parveen and S Palaniappan and R Awang and R Wu and W Peters and M Morgan and B Boser and I Guyon and V Vapnik and G Graksn and H Hakl and H Uuz and C Burges},
  title = {{Comparative Study of Data Mining Techniques on Heart Disease Prediction System: a case study for the}},
  journal = {International Journal of Science and Research (IJSR)},
  year = {2016},
  doi = {10.21275/v5i5.nov163704},
}
Figure