A rule discovery algorithm appropriate for electrocardiograph signals

In this paper the problem of discovering rules, which are associations among patterns in the same time series, is considered. A novel algorithm, namely rule discovery algorithm (RDA), appropriate for periodic time series data, like electrocardiograms (ECGs) can be considered, is proposed. The first phase of the algorithm aims to break the sequence (i.e. an ECG) into overlapping, reconfigured length subsequences according to the sampling frequency and the types of ECG abnormalities to be studied. The Pearson correlation coefficient was chosen as the categorization metric, which is independent of the base line shifts and the amplitude scales. At the following phase the categorized sequence was scanned, so the most efficient rules would be mined. The format of those rules is "IF A occurs THEN B occurs WITHIN time T", where A and B are categorized subsequences and T the time duration between A and B. RDA was evaluated on 60 congestive heart failure patients1 ECGs from a home care monitoring database. The mined rules are complementary to the ECGs' plots allowing the physician to test various hypotheses and discover hidden knowledge.