Skip to Main Content
The application of neural networks in the implementation of ECG classifiers has become widespread. Unfortunately due to the lack of scientific evidence many of the choices made in the design of these classifiers are based on trial and error. This paper details an investigation into a statistical approach aimed at reducing the computational requirements when training an ECG classifier. The multiple linear regression method was used to develop a predictor that would indicate at which point training of a neural network should stop. When tested it was found that this genre of predictor exhibited reasonable accuracy and out performed other predictors based on neural network and genetic programming techniques.