Heart failure prediction framework using random forest and J48 with Adaboost algorithms

Abstract

Heart failure is a very serious condition in health sector globally. It has proven difficult and expensive to manage over the years even with some pre-existing prediction models that signal its occurrence. The predictive accuracies of the existing models are below impressive hence the need for better heart failure predictive models. This work developed two heart failure predictive models to contribute to the decrease in the mortality rate due to heart failure as well as assist patients and physicians in managing the condition. The models were Random Forest(RF) and J48 model with AdaBoost. The dataset for the work was collected from the Cleveland Hospital database. It has 13 attributes and 303 instances. The dataset was preprocessed before use and was converted to the CSV format usable in the Waikato Environment for Knowledge Analysis (WEKA) software. The Agile Unified Process (AUP) methodology was adopted in this work the simulator for the work. The Simulator (web-based) was implemented using Python programming language and the Streamlit for python. The result of the models showed a 92.3% accuracy in prediction for the AdaBoosted J48 model and 89.2% for the Random Forest model. The results indicated that J48 with AdaBoost outperformed RF.