Online outlier detection of learners’ irregular learning processes

Distance education using e-learning has become popular in educational situations. One problem is that the instruction strategy tends to be one way, so sometimes learners find it more boring than conventional instruction methods. This chapter proposes a method of online outlier detection of learners’ irregular learning processes using the learners’ response time to e-learning content. The unique features of this method are as follows: (1) It uses the Bayesian predictive distribution, (2) It can be used for small samples, (3) It unifies the methods of various statistical tests using a hyper-parameter and provides more accurate test results than one of the traditional methods alone. (4) It assists two-way instruction using data mining results of learners learning processes. (5) Outlier statistics are estimated by considering both students’ abilities and the difficulty of content. In addition, this chapter proposes an animated agent which provides adaptive messages to the learners using the data mining. Moreover, the system was evaluated and the results showed the effectiveness of the system.