Improving clustering by detecting and removeing outliers based on Euclidean and exponential distance approach

In this paper, a new method for identifying outlier time series based on GARCH model by exponential distance approach is presented in three steps: first fuzzy and hard clustering methods are implemented on time series, then the outlier time series are detected and removed from the dataset. After removeing outlying time series, clustering algorithms are applied for dataset again. The 30 stocks of the top active, lucrative and profitable stocks in the Iranian stock market are used to evaluate the presented methods. By computing the Silhouette and Xe-Beni indexes, the accuracy of the clustering methods are compared and finally, it is shown that by removong the outlier time series, the GARCH model based on the exponential distance approach has the best performans.