Other Applications: Sequential Dependence Modelling and Data Mining - On Statistical Pattern Recognition in Independent Component Analysis Mixture Modelling - page 161

For each student, the corresponding total instance counter of each kind of event
was calculated, and a normalized value (1-100 scale) of student event activity was
calculated as

even activity student ¼ event total instance student
instance maximum event

100 ð 7 : 6 Þ

The student activity data were added as fields to the datawarehouse plus the
average connection time to the web and average grade obtained for each student.
Course achievement was not always required in virtual courses at UPA, so only
1,873 of the 8,909 rows of the datawarehouse had a value for an average grade.
Besides, those data subsets were divided according to whether they were graduate
or regular academic courses.
Once the data were prepared, we applied the non-parametric Mixca algorithm,
which was described in Chap. 3 , configuring it to estimate one ICA (see Appendix
A). The results are presented in the next section.

7.2.3 Results and Discussion

The ICA algorithm was applied in two stages. In the first stage, ICA was applied
directly to the data in order to determine whether the e-learning web activities
were independent by themselves, i.e., searching for those web activities that can be
separated by an ICA algorithm as a source. The second stage consisted of applying
ICA on a reduced data dimension of five components in order to associate those
components with Felder's learning dimensions.
Figure 7.6 shows the data of the web activities and average connection time and
grades for graduate course with grades. Figure 7.7 shows the sources recovered by
non-parametric ICA of those data. It is a high correlation between event 7
(workgroup documents) and the source s9, and between event 8 (exercise practice)
and the source s5. Therefore, we can assume independence for those events.
After analyzing the results from the ICA applied to the different data subsets,
we can infer the following conclusions:
• Email exchange was independent in some cases. In some courses, e-mail
exchange was not mandatory in the activities.
• The workgroup document event was independent. In the case of courses with no
grades, the lack of evaluation and grades discouraged the participation of stu-
dents in collaborative tasks. In the cases of courses with grades, it was an
optional activity in some courses.
• In some datasets, the content consulting event was independent as a reflection of
the distributed passive learning (DPL) nature of the web platform [ 25 ]. Thus,
content consulting became a routine consisting in download materials with no
interactive learning process.