Extending the Classification Paradigm to Temporal Domains

Abstract

One of the primary areas of machine learning research has been supervised concept learning given some information about examples whose class is known, the goal is to produce a classifier which can classify examples whose class is not known. In general, research in this area has focused on situations where an object’s attributes do not change in the short term. However, in many real-world domains, such as speech, sign language, robotics and medicine, many of the classification tasks involve dynamic attributes. Furthermore, temporal properties are critical to classification. The current work involves developing a temporal classification learner that works in a variety of domains, does not require excessive amounts of data and is able to produce comprehensible descriptions of the concepts, while still having high predictive accuracy.