Adaptive Techniques for Extracting Mental Activity Phases from Heart Beat Rate Streams
The paper presents algorithms for automatic detection of non-stationary periods of cardiac rhythm during professional activity. While working and subsequent rest operator passes through the phases of mobilization, stabilization, work, recovery and the rest. The amplitude and frequency of non-stationary periods of cardiac rhythm indicates the human resistance to stressful conditions. We introduce and analyze a number of algorithms for non-stationary phase extraction: the different approaches to phase preliminary detection, thresholds extraction and final phases extraction are studied experimentally.
Due to very significant differences between streams obtained from different persons and relatively small amount of data common machine learning techniques do not work well with our data. Thus, we had to develop adaptive algorithms based on domain-specific high-level properties of data and adjust parameters based on the preliminary analysis of the stream, making the algorithms adaptive and thus able to capture individual features of a person.
These algorithms are based on local extremum computation and analysis of linear regression coefficient histograms. The algorithms do not need any labeled datasets for training and could be applied to any person individually. The suggested algorithms were experimentally compared and evaluated by human experts.