Structural Change Detection of Time Series Using Sequential Probability Ratio Test

Time series analysis is used in various fields such as not only in economics but also in pattern recognition, biometrics, and Kansei engineering field. The problem of predicting time series can be classified into three in a practical sense. The first problem is how to make a model for prediction, that adequately represents the characteristics of the past time series data. The second problem is how to correctly detect the structural change of the time series as soon as possible, when the estimated prediction model does not meet the real data. The third problem is how to quickly find the new prediction model to meet the real data after the structural change. This paper focuses on the second problem and proposes a method based on a probability ratio test that has been used in the field of the quality control. This paper also shows some experimental results comparing with a conventional method, and presents the effectiveness of the proposed method.