Data Analysis In Data Services Monitoring

Through the analysis of data services, and based on the understanding of some kinds of business exceptions, this paper proposes different monitoring methods for different kinds of business. The main research works are as follows:First, this paper brings time series analysis into business activity monitoring. The innovation of this paper is data preprocessing. This paper detects outliers and change points before the use of ARMA modeling, and then deals them respectively, thus improves the accuracy and reliability of prediction.Second, this paper puts control charts and change point detection into daily business activity monitoring, and decreases false alarms by improving both methods.Third, control charts and change point detection prone to bring in false alarms because of the cyclical nature of the business. So based on the decomposition of time series, this paper just monitors long-term trend in order to increase the accuracy. We can catch any abnormalities, and at the same time, significantly reduce false alarms by making mutual use of time series, control charts and change point detection.Forth, this paper analyzes a large number of alarms, and makes use of correlation coefficient for filtering junk alarms.At the same time, the author has a preliminary understanding of classification which is commonly used in data mining, and poses a new standard which is worthy of further research to measure the quality of a type of classification.At the end of this dissertation, the author also gives some suggestions to the further research in these fields.