FA-1D-CNN Implementation to Improve Diagnosis of Heart Disease Risk Level
Mohammad Mahbubur Rahman Khan Mamun, A.T. Alouani
Abstract
During the last decade heart disease become the leading cause of death around the world. Improving the accuracy of detection of heart disease from readily available biomedical data will enhance possibility of early treatment and low mortality rate. This paper proposes a heart disease diagnosis system using feature optimization algorithm from firefly algorithm (FA) which is a nature inspired swarm technology and a deep learning technique called convolutional neural network (CNN). The automated diagnosis overcomes the problem of nonstationary and nonlinearity nature of ECG wave. FA performs better by finding the global optima faster than other contemporary nature inspired algorithm such as: genetic algorithm or particle swarm optimization. The method was trained and tested using two separate clinically available electrocardiogram (ECG) databases against other machine learning algorithm. The correctly classified outcome using FA-CNN is 88.25% with kappa statistics of .703, while 84.26% correctly classified outcome and kappa statistics of .63 was achieved using same approach without using FA.