Unsupervised Binary Classification of Heart Diseases Using an Autoencoder Model with Boosting Algorithm

Heart diseases are prevalent and encompass a range of cardiovascular disorders with significant health implications and are a leading cause of mortality worldwide. Early detection and effective management are vital in reducing the impact of heart diseases and improving patient outcomes. Traditional approaches in heart disease classification have limitations as they heavily rely on labeled data for training. Obtaining labeled data for training heart disease models is challenging due to privacy concerns and time constraints. This paper presents a new approach to heart disease classification using an unsupervised learning methodology. We employ an autoencoder model with a sigmoid-activated neuron in the last layer of the encoder part and the extracted feature from encoder part is splitted into two clusters based on a threshold value, representing whether or not cardiac disease exists. The unsupervised output of the autoencoder is then fed into the Adaboost algorithm, where misclassified data weights are adjusted iteratively. To ensure generalization across different sections, we train our model using a combined dataset comprising four datasets: Cleveland, Hungarian, Long Beach VA, Switzerland (CHLS dataset) and test on another dataset: Stalog (Heart) Dataset. The outcomes of our experiments validate the efficacy of our approach in obtaining high classification accuracy rates, offering a potential solution for early diagnosis and treatment interventions for heart diseases.