ABSTRACT
An enormous amount of data is produced in healthcare industry which unfortunately is not utilized completely for constructive decision making. We can create effective analysis and classification tools and use this available data to detect and predict diseases. In this paper, we focus on predicting heart disease as well as its severity. A model which uses optimized deep belief network is designed to predict heart disease using medical data such as chest pain type, blood pressure, cholesterol, blood sugar etc. Deep belief network, which is a part of deep learning, is a graphical model composed of multiple layers of hidden units and learns from a very large dataset. Although the function is meticulous and better than many classification models like neural network, but due to large number of input parameters required, it becomes difficult to train and to find the best possible optimized model. In this project, we have optimized deep belief network with a nature inspired algorithm called particle swarm optimization to find the best suited dimension of the restricted boltzmann machine layer and the perceptron layer. Further, to test the proposed strategy, standard UCI Hungarian and Cleveland heart datasets are used. The accuracy came out to be around 90% for Cleveland dataset and 94% for Hungarian dataset.
