Variance based data fusion for k-means++

摘要

The need of clustering the data has been increased day by day in various applications such as Intrusion Detection System, Image Recognition System, etc. Clustering is very much useful in splitting the huge unlabeled data itemset into meaningful groups using similarity metrics. But, at the same time, the cost of the clustering algorithm is computationally expensive for such high dimensional data. Therefore, in our proposed `Variance based Data Fusion K-Means++' data (attribute values) available in multiple dimensions are fused to single (or) very few dimensions. By fusing the data, the vital characteristics of the data are perfectly retained by appropriately weighing the attributes. Proposed preprocessing method is used with K-Means++ and tested with publicly available datasets. From the experiments, it is evident that the VDF K-Means++ achieves high accuracy with fewer false alarms and less processing time than the existing algorithms.