Time series analysis of ECG data and unreliable forecasting

It is of great interest to the medical community if we can predict the fatality of cardiac arrest so that they can provide effective treatment by observing individual Electrocardiograph (ECG) time series data. We usually use linear models because they are simple and easy to apply. But linear systems that are used to describe complex biological system such as ECG data are no longer satisfactory. For the ECG data , it is expected that a non linear system will provide more complete and parsimonious description of the dynamics as well as it will give additional insight into the underlying physiology. But many authors have found that some proposed classical statistical non linear models are totally imable to forecast with reasonable accuracy and that the forecast is even worse than linear prediction. The question naturally arises whether the ECG data at all represent some well known statistical non linear system. In this thesis , we want to understand this problem. For this purpose , we study different sispects of ECG data, which includes understanding the ECG system and ascertaining what specific assumptions could be justified given the nature of the data. This type of problem is widespread in statistics and we shed some more light in this matter from our study.