Enhanced Hoeffding Anytime Tree: A Real-time Algorithm for Early Prediction of Heart Disease

Abstract

Healthcare studies prove that heart disease has increased in recent decades and the growth of patients suffering from heart problems does not stop. In this regard, various data mining techniques have been used by machine learning researchers to support health professionals in the decision-making of this disease. Many of these techniques are based on basic machine learning classifiers, others integrate these classifiers in streaming systems in order to accelerate the execution time. However, some heart situations demand early detection to reduce the chance of having a dangerous illness and the existing machine learning solutions are not appropriate for real-time analysis, because we need to accelerate the algorithms themselves. In this paper, an online algorithm called Enhanced Hoeffding Anytime Tree (EHATT) is proposed to efficiently predict heart disease. EHATT is based on Hoeffding Anytime Tree (HATT), the last version of incremental decision trees. The amelioration that was made by EHATT on HATT, is the change of the node splitting evaluation function with another more suitable for split measures. To examine the performance of EHATT, four metrics are evaluated: classification accuracy, time, memory, and tree size. The experiment results show that EHATT achieves good performance to predict heart disease.

Improvement of Data Stream Decision Trees

“…Different works have taken advantage of the speed and efficiency of EFDT to solve classification problems in various fields, such as (Khine et al (2020), Benllarch et al (2021), Khairi et al (2021)).…”