The impact of feature selection models on the accuracy of tree-based classification algorithms: heart disease case
Yiğit Efe, Leyla Demir · Procedia Computer Science · 2025
Heart disease is a very serious illness that can result with death if it not detected in time. Early diagnosis and effective treatment of heart disease can prevent instead of the progression and improve the quality of life. In recent years, machine learning techniques are widely used to accurately predict the heart disease accurately. One of these techniques is tree-based classification algorithms that construct a tree-like structure while making predictions. Feature selection models are used to improve the performance of machine learning techniques and their ability to generalize by identifying the most informative features. However, there is limited research on how different feature selection models specifically impact the performance of tree-based classification algorithms in the context of heart disease prediction. In this study, the impact of feature selection models on the classification performance of tree-based algorithms to determine the risk of heart disease is investigated. For this purpose, five different feature selection models are applied to the dataset taken from UCI Machine Learning Repository, and the classification performances of eleven different tree-based algorithms are analysed. Classification results show that the Hoeffding Tree technique achieved the highest accuracy (0.84) on the dataset where the stability selection model is applied.