Prognostic Modeling in Cardiovascular Diseases: Unraveling Survival Odds Through Machine Learning

Abstract

Cardiovascular disease (CVD) stands for a class of diseases that involve the heart or blood vessels. It is an umbrella term that contains various conditions affecting the cardiovascular system. Heart failure (HF) is one of the most common CVDs that can impact people of any age. HF is a condition where the heart is unable to pump blood effectively and the body cannot meet the needs of blood. This condition leads to other symptoms like shortness of breath, fatigue, and fluid retention. HF is not a sudden stop of the heart, but rather a chronic condition that develops over time. Even though HF is a major risk factor for patient survival, the presence of additional co-occurring pathological diseases might seriously jeopardize patient outcomes. Heart specialists may find it challenging to estimate a patient's odds of life in HF without the use of computational techniques, which might ultimately lead to the patient not receiving the proper care. This is because several factors affect a patient's chances of survival. Cardiologists can avail the usefulness of machine learning (ML) algorithms to design the proper treatment plan using relevant medical data. The purpose of this research is to develop a prediction model for patient survival in HF conditions depending on several factors which can be interpretable to health practitioners. The predicted result will help the cardiologists to take the necessary steps sooner than the conventional method. In this project, we utilized the UCI HF dataset to develop and evaluate our model. The dataset contains relevant medical information of 299 HF patients. We analyzed and visualized the data to learn about the data quality and distribution. Performed necessary cleaning and feature engineering to get better accuracy in prediction. Our prediction models used the following machine-learning techniques - Naïve Bayes (NB), Logistic Regression (LR), Random Forest (RF) Classifier, XGBoost (XGB) Classifier, and LightGBM to find accurate results. This paper also presents how we ensured the accuracy of the outcome using hyperparameter tuning and avoiding overfitting. We also focused on the precision score, recall score, f1 score, and area under the curve (AUC) rather than only focusing on the Accuracy score. Our research shows that the RF Classifier performed best with an Accuracy of 87.78 % and an AUC of 94.35% in predicting patient survival in the case of HF. However” LhihtGBM reached the Accuracy of 88.89%.