Optimized Feature Engineering and Machine Learning Methods for Heart Disease Prediction

Heart disease is a chronic global health concern that accounts for the highest death rates, thereby necessitating prompt medication and treatments. With the aim to enhance the prediction, this paper proposes a machine-learning model that analyses patient medical parameters to detect heart diseases and their severity. This study uses 6 machine learning algorithms and feature engineering methods StandardScalar, SelectKBest and Principal Component Analysis in a pipeline grid that reduces the computational overload. The Kaggle Heart Disease open access database is thoroughly cleaned to remove 25 erroneous values which helps understand the underlying patterns in the patient database. Feature engineering pipeline grid develops a new feature set with GridSearchCV for hyperparameter tuning. Parallel processing Joblib library “parallel” and “delayed” functions speed up the performance and the processing time of the classifiers. This model demonstrates better performance with Gradient Boosting achieving 100% accuracy and faster processing time of 0.003ms that is better than the existing models.