Abstract:

Today Massive Open Online Courses (MOOCs) are actively used in both higher education, and continuing vocational education systems. The reasons for that include relatively low cost of online learning, its high accessibility, and wide spectrum of opportunities provided. At the same time, online education has a number of shortcomings arising out of the lack of contact with a teacher, and low level of students’ self-motivation.

Success rate of online course students depends upon the number of factors, including content and assessment materials’ quality, individual specifics of course apprehension by the students, and the level of support provided for students in course of learning. Learning analysis tools allow studying behavioral patterns of the students, and forecast the rate of success based upon analysis of digital footprint left on the open education platform.

The paper presents two models for intellectual data mining:
1. Probability model for group forecasting based upon the theory of Markovian processes. The model allows forecasting statistical distributions for student final testing grades based upon the current academic performance data.
2. Statistical model for individual forecasts using traditional data mining methods of clustering and regression. The model allows personal forecast of student final testing success rate based upon current performance data accumulated during the course.

The former models were implemented and tested for “Engineering Mechanics” course of the Ural Federal University. The course is hosted on the National Open Education Platform (openedu.ru).

Data sources included:
- Grade Reports, covering student grades obtained during the course, and final tests;
- Log files of an educational platform generated when students were taking online courses.

Modeling results include the success rate forecast for a student taking a final test, based on current grades of student. Forecasts were made individually for all active course participants after each milestone of a course.

Resulting forecasts can be used to provide personal support for MOOC students, perform timely corrections of observed negative trends in academic dynamics, and develop individual recommendations and educational path for each student. Results would allow combining mass involvement that is characteristic for MOOCs with individual learning approach specific for traditional education models.

Keywords:

Massive Open Online Courses, learning analytics, intellectual data analysis models, forecasting academic results.