GIFT

Abstract: The grand challenge of Intelligent Tutoring Systems (ITS) development is that of creating a computer tutor as good as a human tutor. This difficult task may be broken into several parts. The first task real instructors perform prior to making instructional decisions is assessing the state of the trainee. Thus, the first consideration in the construction of an ITS is obtaining meaningful data from sensors and interpreting them in order to assess trainee emotional state. The interpretation of sensor data is the significant problem in this area, with the problem of sensor data mostly having been reduced to sensor selection. The machine learning methods for interpreting unlabelled sensor data are significantly more sparse than the sensors available, and their selection is far from straightforward. In this paper, Growing Neural Gas (GNG) methods, two types of incremental clustering, and Adaptive Resonance Theory (ART) will be evaluated against each other on fabricated and realtime data streams of trainees‟ state in order to determine the best selection of methods to accomplish this task.