Deep Learning Based Decision Support System

Abstract

Heart diseases are amongst the leading causes of death worldwide. Death becomes certain if heart diseases are not diagnosed in the early stages. It has been causing death irrespective of age, gender or any other demographics. That is why diagnosing heart disease is extremely important. The most commonly used clinical method for diagnosing heart disease is angiography which is a very expensive procedure. Studies have also suggested that it has negative side effects. Lately, a lot of research has been conducted based on the use of machine learning for heart disease diagnosis. This thesis has also conducted a systematic literature review in order to thoroughly analyze the existing literature and look for gaps in it. The results of the review reveal that the most popular classification techniques are Support Vector Machine, Neural Networks, and ensemble classifiers. The main objective of this thesis is to present a technique that can be used to develop a decision support system or a computer-aided diagnosis system for heart disease particularly coronary artery disease. Studies reveal that the performance of neural networks can be increased by systematically initializing attribute weights instead of random weights initialization. In order to attain a reliable methodology, four feature selection i.e. weights by Support Vector Machine, weights by Principle Component Analysis, weights by Gini Index and weights by Information Gain and four weight optimization techniques i.e. Forward, Backward, Particle Swarm Optimization and Evolution Strategy have initially been used to provide optimized attribute weights in order to improve the performance of artificial neural network. The results of the initial experimentation are promising. Later, an ensemble is formed by getting an average attributes weights of three weight optimization techniques i.e. PSO, ES and backward weight assignment. The average weights are then provided to the input layer of the neural network. The accuracy attained by the proposed system is over 94%, which is promising. In future, the proposed technique can be used to form a reliable and assistive system which can be used as a diagnostic tool in order to add clinicians and physicians.