Extracting adaptation strategies for e-learning programs with XCS

This paper investigates XCS performance on a scarce and noisy artificial and a real-world data set. The real-world data set is derived from an E-Learning study, in which motivation was correlated with the adaptation of difficulty. The artificial data set was generated to evaluate if XCS can be expected to mine information from the real-world data set. By adding sparsity and noise to the artificial data set, mimicking the properties of the real-world data set, we show that XCS can handle scarce and noisy data well. We furthermore show that the extracted structure contains problem-relevant information, and that revealed structures in the real-world data correspond to actual psychological learning theories. Thus, the contributions of the paper are twofold: (1) We show that XCS can mine highly scarce and noisy data; and (2) the results suggest that the current motivational state of the user may be utilized to adapt an E-Learning program for improving learning progress.