Using Decision Trees for Discovering Problems on Adaptive Courses
Javier Bravo1, César Vialardi2 and Alvaro Ortigosa1 1Computer Science Department, Universidad Autónoma de
Madrid, Spain2Computer Science Department, Universidad de Lima, Peru
{javier.bravo, alvaro.ortigosa}@[email protected]
IndexImproving an adaptive courseStructure of logsData for the experimentsAnalysis of the dataFirst experimentSecond experimentConclusions and future work
Improving an adaptive course
Students
Instructor
User Model
Authoring ToolCourse Delivering System
Student behavior
Student results
Student paths
<log><profile>
name=“John” age=“12” experience=“normal” </profile>
<entry>activity=“eo1_n1” activityType=“P”complete=“1.0” grade=“1.0”numvisits=“1”timestamp=“2005-12-14T11:19:50.879+01:00”type=“LEAVE-ATOMIC” </entry>
</log>
Structure of logs
Level of completeness of the
activity
Score in the activity
Time when the student visits the activity Number of times the
student has visited the activity
Action executed by the student
Profile of the student
Type of the activity
Name of the activity
Data for the experimentsStudents:
24 students.Age between 12 and 14. First year of secondary mandatory education.
Adaptive course:Introduction to whole numbers.Seven lessons, 22 practical activities.Two levels of adaptation: novice and normal.
Analysis of the data
First experiment
Objective: to find potential problems in the adaptation.
Steps:Select the practical activities of the logs.Build a decision tree:
Attributes: age, experience and activity. Classification attribute: success.
Analyze the decision tree: searching from the leaves with not success to the top.
Results of first experimentactivity
yes (23/6) age
yes (6/2)no (18/8)
<=12 >12
no (24/6)yes (24/3)
=er1_b1=ev1_n1=eo1_n1 =es2_n1
no (24/3)
=em2_b1
no (22/1)
=ec3_a1
no (28/4)
=ep1_b1
Second experiment
Objective: to find accurate information about the potential problems in the adaptation.
Steps:Analyze the proportion of failures for different
profiles of students.Simulate 100 students with these proportions
of failures by using Simulog.Build a decision tree.Analyze the decision tree.
Results of second experimentactivity
=ep1_b1
yes (54/17)
no (56/14) yes (15/6)
<=12
age
yes (8/3)no (26/12)
<=12 >12
experience
=novice
yes (12/2)
=normal
age
>=12
yes (81/28)yes (55/19)
=ep1_a1=er2_b1=er2_a1
yes (55/19) no (82/22)
=ec3_a1=ec3_n1=em2_b1
Profile Activity
Age=12Experience=Normal
ep1_b1
Age=12 em2_b1
All ec3_a1
ConclusionsThis work shows the utility of using data
mining methods with real student data.The first experiment obtained less
information of profiles of students with problems.Is related this lack of information with the size
of data set?The second experiment obtained accurate
information of profiles of students with problems.The size of data set influences on the
information provided by the decision tree.
Future work
Support the results of decision trees with other learning methods: associations rules and clustering.
Developing a tool for assisting instructors on understanding the results provided by decision trees.
Questions
Top Related