A THESIS SUBMITTED TO THE SCHOOL OF GRADUATE STUDIES IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE DEGREE

Abstract

Cardiovascular diseases are prevalent in low- and middle-income countries, including Ethiopia, where they are the primary cause of public health issues. Despite the high prevalence of cardiovascular diseases in Ethiopia, many cases remain uncontrolled due to insufficient risk factor systems for efficient data preparation. Addressing this issue is vital for advancing public health initiatives and ensuring that appropriate interventions are implemented on time.. Additionally, proper disease diagnosis early on allows for timely countermeasures before the disease spreads. These advancements are crucial for improving health outcomes in the region. The total amount of dataset collected by researcher is 1,981 records with twenty-two attributes from 2017-2024 G.C, obtained from Arbaminch General Hosipita. The researchers used basic risk factors like electrocardiogram, respiratory rate, pulse rate, and oxygen saturation percentage. A study was conducted to develop explainable cardiovascular disease prediction using a machine learning techniques. The main goal of this research is to explore the topic and address the existing research gap by experimentally using a machine learning techniques to develop explainable cardiovascular disease prediction. In this, the researcher used different feature selection methods to produce feature importance for early cardiovascular disease prediction model outcomes like mutual information, random forest, shap and lime methods. The researcher used machine learning models such as CatBoost, XGBoost, and LightGBM with SHAP and LIME methods. To train and test the model we selected catboost for an interpretable model. Performance metrics such as accuracy, precision, recall, and F1 score were evaluated. The CatBoost model outperformed than all other models, with 99.7% accuracy and 100% ROC AUC