<![CDATA[ABSTRACT -- Many researches show that time series data is a kind of data which has biggest volume and growth compared to the others kind of data. Parallel with its huge volume and rapid growth, it is always needed new method, technique or approach to explore knowledge contained in time series data. One of many goals in data mining of time series is anomaly detection. By definition fractal is an object that has such self similarities in certain degree. This paper presents the results study of HFD (Higuchi Fractal Dimension) approach for clustering to detect the existence of an anomaly or deviation in time series data. This proposed method is applied to PT.PGN daily stock trade year 2004 to 2012 as test data. The results show that for value of discrete interval k = 5, 10, 15 their HFD tend to diverge and there are tend to converge to 2 for the greater value of k. Based on HFD time series clustering results, this approach can be used to divide the normal data and the other data that contain certain anomaly. In this study is also shown that this approach provides a better result compared to adjustment method which fill unbalance time series data with a zero value. Keywords: time series, anomaly, dimensionality reduction, fractal dimension, clustering]]>
Copyrights © 2013
