Study On Change-point Detection Method Based On Hierarchical Clustering Analysis

Time ordered data widespread in many fields. And it is often used to evaluateprocess performance. Processes often change due to kinds of elements over a longperiod of observation which results in the difference performance of the products.Many factors such as raw materials and equipment changes can be contributors.Identifying accurate time that data change can help us analyze the causes for variationand detect any changes in a process in order to improve it. The aim of this paper is toidentify change-points effectively. Change-points are the corresponding accurate datapoints when the process changes.One of the important characters of industrial history data is ordered by time.Industrial process tends to have a stable performance before transferring to anotherstable performance. This type of data is called time-ordered data. Indentifying thechange-points of time ordered data accurately is helpful to analyze and find thereasons of process shifts and improve the industrial process.This paper presents a combining method of hierarchical clustering algorithm andhypothesis testing for unvaried time ordered data clustering and detecting changepoints. And by comparison with the existing methods on performance, it is provedthat the method in detecting change points is more efficient and accurate. This methodis divided into two parts. The first part is a combining method of divisive hierarchicalclustering and Z test. The method divides the data into several parts and indentifiesthe change points. The second part is a combining method of agglomerativehierarchical clustering and two sample T test. The method agglomerates the divideddata and removes the outliers indentified in the first part. And this paper presents adeep research on different unvaried time ordered data and different hypothesis testused in the method by comparing the related performance indicators. By analyzing thedeep research the paper summarized the advantages and disadvantages of the methodand finds the areas for improvement.