Holter electrocardiogram has spread with medical institutions widely to be able to detect transient arrhythmia and ischemia. It is necessary to be analyzed automatically, because of a large quantity of data that is recorded for hours. However, the accuracy of automated classification for QRS complexes is not sufficient. This paper presents the application of a hierarchical self -organizing map (SOM) to classify QRS complexes. Each beat is divided into some sections, and the characteristics are learned by the first SOM. The qualitative attributes are classified for characteristics of each section. The qualitative attributes are arranged in time series for each beat, and they are learned in the second SOM. As a result, each beat is classified by their characteristics. We evaluated this method using MIT-BIH Arrhythmia Database of 17 cases 33,362 beats and compared to a correlation coefficient method. The classification error rate was 0.41% and the number of the classifications was 28.7. Using this method the classification of QRS complexes is significantly improved.
