Seminar 61--Outliers Detection Techniques and their Benefits in Data-Driven Modeling

This product is a zip file that contains files that consist of PowerPoint slides synchronized with the audio-recording of the speaker, PDF files of the slides, and audio only (mp3 format) as noted.

An outlier is an observation which deviates from other observations and arouses suspicion. Monitored data are inherently prone to errors caused by instruments malfunction, power surges, disturbance and noise especially with wireless sensors, among other causes. Monitored building energy use data contain anomalous values and gaps of missing readings. The session covers the outliers detection in data-driven modeling, discussing approaches and techniques, including statistical inference, data mining, and artificial intelligence methods, along with an explanation of types of outliers. Cases studies are presented and cover baseline modeling of building energy use, thermal comfort models and whole building energy data screening.