Title: Outlier detection using auxiliary variable dependent monitoring in scanner data
Authors: Youngrae Kim - Seoul National University (Korea, South) [presenting]
Abstract: Recently, statistical agencies of major countries analyze the scanner data, which include the information about transaction, such as price and quantity. It can be important to detect sudden change or anomalies in transactions. We turn this problem into outlier detection problem in price change rates. Because the existing methods use only the transaction price information, there is a difficulty that can not deal with the situation with the abnormal sales amount. We develop the procedure to detect the anomaly of the transaction by considering both the transaction price and quantity information. We estimate variance of price change rate using quantity information with kernel estimation method, to create abnormal point detection criteria. And we can show the criterion on the control chart. We divided the two periods, estimation period and monitoring period, and propose appropriate length of estimation period through a simulation study. We numerically compare the performance of our method to existing methods which did not use volume information.