Anomaly Detection by Using Streaming K-Means and Batch K-Means

This paper introduces K-Means algorithm as new technique for detecting anomaly. Data analysis has been applied to industry field widely and plays important role in it. However, conventional data analysis method cannot process large-scale data in considerable time and waste lots of computing resources. Conversely, Batch processing and Stream processing are equipped with property of processing data in short time interval, especially stream processing, can process data in real-time. This paper also compares Batch K-Means processing with Streaming K-Means processing according to distance, cost value and cluster distribution factors. Moreover, this paper also discusses how to reach optimized K value of Batch K-means model and Streaming K-means model, analyzes attributes of Batch K-Means processing and Streaming K-Means processing and finds limitations of these two processing models. Finally, the paper proposes limitations of research experiment and future improvement of clustering technique.