| Abstract [eng] |
The technology of today allows us to monitor various processes for a long time. One of such examples are the monitoring signals obtained from human wearing sensors. Such sensors enable users to capture and accumulate the human heart, respiratory and other related process signals for a long time. Therefore, the creation of methods and techniques that can handle large volumes of data and enable early detection of a variety of health-related changes, becomes a relevant task. This paper addresses application of long-time monitored processes parameters in classification task to witch it is offered a methodology that allows the classification to use the calculated parameters of the Detrended Fluctuation Analysis method. For methodology implementation purposes three programs were created: MATLAB function for application of Detrended Fluctuation Analysis and assessment of method appropriateness for the data, SAS macro for analysing self-similarity parameter differences between the groups, and R code for classification trees implementation and comparison. Proposed methodology was applied to electrocardiogram’s RR interval sequences to detect congestive heart failure. Analysis has shown, that Detrended Fluctuation Analysis method was appropriate for the data. Means of two self-similarity parameters shown significant differences within the groups. The C5.0 algorithm classification tree, based on self-similarity parameters and age attribute, shown the best fit for analysed data and classified 91% of the data correctly and 83% of the congestive heart failure data correctly. |
