Plenary Lecture

Change Point in Time Series Data

Professor Azami Zaharim
Coordinator for the Unit Fundamental Engineering Studies
Faculty of Engineering and Built Environment,
Universiti Kebangsaan Malaysia,
43600 UKM, Bangi, Selangor
MALAYSIA

Email: azami@eng.ukm.my

Abstract: In building a statistical model for time series data the primary concern is to know whether all the observations can be represented by one particular model or whether the parameters in the model change at some known or unknown time point, called the change point. Subsequently, change points are defined as the points in data where two adjacent segments of the time series are connected. However, there are real-world applications in which only the position of the change is required and not the fitting functions. A change point can occur as a change in mean; change in variance or covariance or both; change in parameter; change in the structural model; or change in the trend in the model at certain known or unknown time point. Time series change point can be classified into two main categories; those which infer in a change when the statistics exceed a control limit; and those which directly estimate the time of change. In each category, the time point is a main factor, where the construction of the statistics and estimation are based on whether the time of occurrence is not known or not. Practically in most cases the time of change is unknown. From the simulation, it can be conclude that the larger the difference of the parameter estimates before and after the given change point, the higher will be the probability of the detection of the change point, the models that do not include a regular differencing operator, tends to be slightly higher in the probability of detection than the others, similar results occur for seasonal and non-seasonal models but the detection for the change point will be slightly lower for the seasonal models, and the procedure does not perform well when the point of change is at the beginning or at the end of the series.



Brief Biography of the Speaker:
Azami Zaharim worked first 13 years as a lecturer in the Universiti Teknologi MARA (University of MARA Technology - UiTM) before joining the Universiti Kebangsaan Malaysia (National University of Malaysia - UKM) in the year 2003. He is Associate Professor at the Faculty of Engineering and Built Environment UKM, and is currently Coordinator for the Unit Fundamental Engineering Studies. He obtained his BSc(Statistics and Computing) with Honours from North London University, UK in 1988 and PhD (Statistics) in 1996 from University of Newcastle Upon Tyne, UK. He specialize in statistics, public opinion, engineering education and renewable energy resources.
He has until now published over 80 research papers in Journals and conferences, conducted more than 15 public opinion consultancies and delivered 3 keynotes/invited speeches at national and international meetings. He is currently the head of Renewable Energy Resources and Social Impact Research Group under the Solar Energy Research Institute (SERI). In the year 2007, he headed the Engineering Mathematics Research Group. At the same time, he is currently active involve in outcome based education (OBE) approach at the national level and the chairman of the Engineering Education Research Group since 2005. He is also involved actively in the research for the future of engineering education in Malaysia 2006 under the Ministry of Higher Education of Malaysia.