Nonlinear ARX modeling of heart diseases based on heart sounds

This paper proposed the heart disease modeling system based on heart sounds. The model uses ARX model as regression vector and Neural Network as nonlinear model structures. The number of hidden neurons was optimised by minimizing the criterion of NSSE, fit and FPE criterion. The model architecture of 2-4-1 perfectly fits the original heart sound signals with average R-square of above 99.9%. The weight parameters of the models were then estimated and analysed for the purpose of classification of the heart diseases.