Prof. He-Qing MU
(Department of Civil Engineering, South China University of Technology, Guangzhou, China)
Title
Outlier detection and post-analysis
Abstract
In this seminar, we focus on outliers in data analysis and statistics.
In practical circumstances, it often occurs that some data points deviate substantially from the model output.
The presence of outliers indicates irregularities of data and/or weakness of the model/theory. The study conducts a two-stage analysis of outlier.
In the first stage, outlier detection is implemented.
Some outlier detection methods are reviewed and a probabilistic measure-based outlier detection criterion is proposed.
In the second stage, a classification algorithm is introduced to conduct post-analysis of outlier based on the results in the first stage.
Finally, a concept of ‘prediction unreliable region' is proposed to quantitatively show the weakness of the assumed mathematical model.