December 2012

Classification in Data-Rich Dynamic Environments

Nicos Pavlidis, Lancaster University Management School

Venue: A54, Postgraduate Statistics Centre, Lancaster University LA1 4YF

Date: 06-12-2012


Progress in computer technology is radically changing every aspect of our economic and social system. Until recently the focus has been on the increasing speed of computer processors, captured by Moore's Law. Nowadays, the focus has shifted to data generation and storage technology.

It is becoming increasingly recognised that we are surrounded by data. Advances in sensing and storage technologies are fuelling a virtual explosion in the quantity, quality, and variety of data that is becoming available, giving rise to what is known as Big Data. This development is of fundamental importance as it enables us to answer questions which would have defied an answer only a few decades ago. However, the closer computer technology enables us to record real life data generating processes the more the complexity of these mechanisms is reflected in the data; revealing problems that previously could be conveniently disregarded. Some of the key challenges when dealing with Big Data are high volume, high dimensionality, high frequency, and time-variations in the data generating process.

This talk will discuss classification problems when the data originate from a time-varying process and observations are obtained sequentially and at a high frequency. Applications include credit scoring, fraud detection, medical imaging, and high frequency foreign exchange data.