Educational Data Mining

1
Educational Data Mining

• 2008. 9. 17
• Junhyup Song
• Data Mining Laboratory, SNU

2
Intelligent Tutoring System

• Intelligent Tutoring System
• Any computer system that provides direct
  customized instruction or feedback to students
  without the intervention of human beings while
  performing a task.
• The Structure
• Interface model.
• Expert(Domain) model.
• Student model.
• Tutor(Teaching) model.

3
Intelligent Tutoring System

• ITS Conference
• Held in Montreal(2008), Taiwan(2006), Brazil(2004
  ), and San Antonio(1998).
• Educational Data Mining 08
• The first international conference on EDM.
• The increase in instrumented educational s/w has
  created large repositories of data.
• Focuses on computational approaches for using
  data to address important educational questions.

4
Educational Data Mining 08

• Session A Assessment
• Adaptive Test Design with a Naive Bayes
  Framework.
• Session B Improving Skill and Domain Models
• Labeling Student Behavior Faster and More
  Precisely with Text Replays.
• Session C Improving Understanding of Student and
  Tutor Behaviors
• Analytic Comparison of Three Methods to Evaluate
  Tutorial Behaviors.
• Session D Tools to Support EDM
• Data Mining Algorithms to Classify Students.

5
Session A Assessment

• Adaptive Test Design with a Naive Bayes Framework
  1
• Motivation
• The Bayesian model is a statistical model of an
  examinees chances of success or failure based on
  previous observations.
• It is impractical to administer a test of too
  many questions to examinees.
• Objective
• How to optimize the choice of test items for
  collecting the data that will be used for
  training a Bayesian model.

6
Session A Assessment

• Adaptive Test Design with a Naive Bayes Framework
  1
• DM Implementation
• Real datasets from tests in four different
  domains math, UNIX, Arithmetic, French.
• Used the Naïve Bayes Classifier.

7
Session B Improving Skill and Domain Models

• Labeling Student Behavior Faster and More
  Precisely with
• Text Replays 2
• Motivation
• Models classifying student behavior, such as
  gaming the system and help avoidance have been
  developed.
• Developing classifiers of student behavior is
  either highly time-consuming or difficult to
  validate.
• Objective
• Quickly and accurately labeling data in terms of
  student behavior.
• Text replays represent a segment of student
  behavior from the log files.

8
Session B Improving Skill and Domain Models

• Labeling Student Behavior Faster and More
  Precisely with
• Text Replays 2
• DM Implementation
• Used DT to predict whether each student action is
  gaming.
• Exploiting properties or learning the material.
• Cognitive Tutor Algebra from the Pittsburgh
  Science of Learning Center DataShop. Algebra I
  2005-2006(Hampton only)
• https//learnlab.web.cmu.edu/datashop/
• 436,816 student actions(answering, requesting
  help, etc.) by 59 students.

9
Session C Improving Understanding of Student and
Tutor Behaviors

• Analytic Comparison of Three Methods to Evaluate
  Tutorial
• Behaviors 3
• Motivation
• Various methods to analyze fine-grained tutor
  data in order to evaluate the effects of tutorial
  actions on student behavior.
• Objective
• Comparing three different methods empirically.
• DM Implementation
• Randomized controlled trials analysis.
• The participants are randomly assigned to receive
  one of several different interventions(teaching).
• T-test, ANOVA

10
Session C Improving Understanding of Student and
Tutor Behaviors

• Analytic Comparison of Three Methods to Evaluate
  Tutorial
• Behaviors 3
• DM Implementation
• Learning decomposition.
• Exponential learning curve.
• Knowledge tracing.
• Dynamic Bayes network(DBN).

11
Session D Tools to Support EDM

• Data Mining Algorithms to Classify Students 4
• Motivation
• Prediction/Classification of a students
  performance is very important in web-based
  educational environments.
• There are different types of algorithms to
  predict/classify student outcomes, marks, or
  scores.
• Objective
• Comparing different DM techniques based on both
  students usage data in a web-based course and
  the final marks obtained.

12
Session D Tools to Support EDM

• Data Mining Algorithms to Classify Students 4
• DM Implementation
• 7 Moodle courses with Cordoba University
  students.
• Moodle data mining tool.

13
Session D Tools to Support EDM

• Data Mining Algorithms to Classify Students 4
• DM Implementation

14
Military Education and Training

• Web Based Education
• Sexual Harassment Prevention Education.
• Regulations or Operational Plan Education.
• Air Traffic Control Training
• Not a web based education.
• Orientation(2 weeks) Simulation(8 weeks) On
  the Job Training (10 weeks)
• Different courses WD, WAO, SWAO, SD
• Core skills of a controller
• Generally individuals with good memory, are
  organized, have spatial awareness, and are
  quick with numeric computational skills.
  (Wikipedia)
• English speaking(radiotelephony), Team Work.

15
Military Education and Training

• EDM Objectives for ATC Training
• Web based tutoring system to save the time and
  efforts.
• Customized training to improves the core skills.
• EDM Issues for ATC Training
• Assessment
• How to design the ATC test.
• Improving Skill and Domain Models
• Situational Awareness.
• Improving Understanding of Student and Tutor
  Behaviors
• How to evaluate tutorial behaviors.
• Classify Students
• How to predict trainees success or fail.

16
References

• 1 Michel Desmarais, Alejandro Villarreal,
  Michel Gagnon, Adaptive Test Design with a Naive
  Bayes Framework, 1st International Conference on
  Educational Data Mining, pp48-56, 2008.
• 2 Ryan Baker, Adriana de Carvalho, Labeling
  Student Behavior Faster and More Precisely with
  Text Replays, 1st International Conference on
  Educational Data Mining, pp38-47, 2008.
• 3 Jack Mostow, Xiaonan Zhang. Analytic
  Comparison of Three Methods to Evaluate Tutorial
  Behaviors, 1st International Conference on
  Educational Data Mining, pp28-37, 2008.
• 4 Cristobal Romero, Sebastián Ventura, Pedro G.
  Espejo, Cesar Hervas. Data Mining Algorithms to
  Classify Students, 1st International Conference
  on Educational Data Mining, pp8-17, 2008.

17
Q A
18
Back up slide

• Situational awareness.
• The perception of environmental elements within a
  volume of time and space, the comprehension of
  their meaning, and the projection of their status
  in the near future.