A Novel Approach for Heart Disease Detection Using Hyperparameter-Tuned Random Forest Ensemble Method

Abstract

Introduction: Heart related disorders relics as foremost reason of mortality worldwide, emphasizing the significant need for accurate and timely detection methods.

Objectives: This work presents a machine learning approach tailored for detecting heart-related disorders, and the random forest algorithm is enhanced with an ensemble learning approach (RF-EM). Within the domain of heart disease detection, the Random Forest technique stands out for its effectiveness, mainly due to its ability to manage high-dimensional datasets and large volumes of data efficiently. Its incorporation of randomness at two pivotal stages - through the random sampling of data points with replacement and the random feature selection at each split - acts as a protective measure against overfitting, a common challenge encountered in traditional decision tree models.

Methods: The RF-EM model is trained on three different datasets — Cleveland, Statlog and Hungarian. The model also goes through a careful hyperparameter tuning to get the best performance before training starts. This intensive methodology empowers the Random Forest Ensemble technique to be more refined and prepared by learning affability in many datasets, which makes it an accurate method for heart disease determination.

Results: The detailed analysis shows that the Random Forest classifier with default hyperparameter setting had an accuracy of 96.31%. Although with the use of hyperparameter optimization techniques, its precision raised to 97.61%. Furthermore, by applying the Grid Search Cross-Validation (CV) method, the Precision is improved up to 97.90%.

Conclusions: The above results clearly shows that the Random Forest Ensemble Method, will report better prediction for Heart Disease.