Data Mining for an Educational Webbased System

1
Data Mining for anEducational Web-based System

• Behrouz Minaei
• Department of Computer Science and Engineering
• Thesis Proposal
• January 30th 2004

2
Topics

• Statement of problem
• Classification (Prediction student performance)
• Clustering (Ensembles of multiple clusterings)
• Additional Proposed work
• Tentative Schedule

3
Statement of problem

• Statement of problem
• LON-CAPA
• Data Mining
• Data Preprocessing
• Contributions
• G. Albertelli, B. Minaei-Bigdoli, W.F. Punch, G.
  Kortemeyer, and E. Kashy, Concept Feedback In
  Computer-Assisted Assignments, Proceedings of
  the (IEEE/ASEE) Frontiers in Education
  conference, 2002 Boston
• M. Hall, J. Parker, B. Minaei-Bigdoli,G.
  Albertelli, G. Kortemeyer, and E. Kashy,
  Gathering and Timely Use of Feedback from
  Individualized On-line Work with an Open-Source
  CMS submitted to (IEEE/ASEE) FIE 2004 Frontier
  In Education, Oct. 2004 Lavannah
• Classification (Prediction student performance)
• Clustering (Ensembles of multiple clusterings)
• Additional proposed work
• Tentative Schedule

4
LON-CAPA

• This research is a part of the latest online
  educational system developed at Michigan State
  University (MSU), the Learning Online Network
  with Computer-Assisted Personalized Approach
  (LON-CAPA).
• Learning Content Management System
• 9 high schools, 2 community colleges, and 17
  universities nationwide
• Assessment System
• Online assessment with immediate feedback and
  multiple tries
• Different students get different versions of the
  same problem
• Different options, graphs, images, numbers, or
  formulas
• Open-Source and Free (GPL, Runs on Linux)

5
LON-CAPA Data

• Three kinds of growing data sets
• Educational resources web pages, demonstrations,
  simulations, individualized problems, quizzes,
  and examinations.
• Information about users who create, modify,
  assess, or use these resources.
• Data about how students use and access the
  educational materials

6
MSU Fall 2003

• 40 courses used LON-CAPA at MSU
• Total student enrollment approximately 3,067 (out
  of 13,400 total global student-users)
• Disciplines included Advertising, Biochemistry,
  Biology, Chemistry, Finance, Geology, Math,
  Physics, Plant Biology, Statistics for Psychology

7
Statement of problem

• LON-CAPA collects data for every single access to
  the resources in both activity log and student
  database
• Logs are not only huge but also distributed and
  specific to a web-based educational system
  (LON-CAPA)
• Intelligent automated tools needed to discover
  relevant, useful, and interesting patterns
• Apply the discovered rules to produce more
  intelligent system

8
Knowledge Discovery Process

• Data Integration, removing inconsistency,
• Data Cleansing, correcting errors, missing values
• Discretization, transform continuous to
  categorical
• Feature Selection, features are more relevant
• Mining process, rule discovery
• Post-processing,
• Large set rules ? simplify
• 1) More comprehensible, 2) More interesting
• Use combination of objective and subjective
  approaches

9
Data Mining Tasks

• Classification
• The goal is to predict the class variable based
  on the feature values of samples Avoid
  Overfitting
• Clustering (unsupervised learning)
• Association Analysis
• Find the binary relationship among the data items
• Any feature variable can occur both in antecedent
  and in the consequent of a rule.

10
Contributions (1)

• Our claim is that data mining can help to design
  better and more intelligent educational web-based
  environment

Can help instructor to design the course more
effectively, detect anomaly
Can help students to use the resources more
efficiently
11

Contributions (2)
12
Contributions (3)
Can find some associative rules between
students educational activities
Can be used to identify those students who are at
risk, especially in very large classes
Can help instructors predict the approaches that
students will take for some types of problems
13
Prediction student performance

• Statement of problem
• Classification
• Combination of Classifiers
• Weighting the features
• Using a Genetic Algorithm to find the best set of
  weights
• B. Minaei-Bidgoli, W.F. Punch, Using Genetic
  Algorithms for Data Mining Optimization in an
  Educational Web-based System, GECCO 2003,
  2252-2263, July 2003 Chicago.
• B. Minaei-Bidgoli, D.A. Kashy, G. Kortemeyer,
  W.F. Punch, Predicting Student Performance An
  Application of Data Mining Methods with an
  educational Web-based System, (IEEE/ASEE) FIE
  2003 Frontier In Education, Nov. 2003 Boulder
• Clustering (Ensembles of multiple clusterings)
• Proposed work
• Tentative Schedule

14
Data Set PHY183 SS02

• 227 students
• 12 Homework sets
• 184 Problems
• 80 MB activity log
• 26 MB useful data
• 220,000 transactions
• Extracted Features

• Total number of correct answers. (Success rate)
• Success at the first try
• Number of attempts to get answer
• Time spent until correct
• Total time spent on the problem
• Participating in the communication mechanisms

15
Class Labels (3 possibilities)
2-Classes
 
3-Classes
 
9-Classes
16
Classifiers

• Non-Tree Classifiers (Using MATLAB)
• Bayesian Classifier
• 1NN
• kNN
• Multi-Layer Perceptron
• Parzen Window
• Combination of Multiple Classifiers (CMC)
• Genetic Algorithm (GA), Optimizer
• Decision Tree-Based Software
• C5.0 (RuleQuest ltltC4.5ltltID3)
• CART (Salford-systems)
• QUEST (Univ. of Wisconsin)
• CRUISE use an unbiased variable selection
  technique

17
Fitness/Evaluation Function

• 5 classifiers
• Multi-Layer Perceptron 2 Minutes
• Bayesian Classifier
• 1NN
• kNN
• Parzen Window
• CMC 3 seconds
• Divide data into training and test sets (10-fold
  Cross-Validation)
• Fitness function performance achieved by
  classifier

18
Individual Representation

• The GA Toolbox supports binary, integer and
  floating-point chromosome representations.
• Chrom crtrp(N, FieldDR) creates a random
  real-valued matrix of N x d, where N is number
  of individuals (200) and FieldDR is a matrix of
  size 2 x d and contains the boundaries of each
  variable of an individual.
• FieldDR 0 0 0 0 0 0 lower bound
• 1 1 1 1 1 1 upper bound
• Chrom 0.23 0.17 0.95 0.38 0.06 0.26
• 0.35 0.09 0.43 0.64 0.20
  0.54
• 0.50 0.10 0.09 0.65 0.68
  0.46
• 0.21 0.29 0.89 0.48 0.63
  0.89

19
Results of using GA
20
GA Optimization Results
21
Features importance
22
Contribution of the classification

• A new approach to evaluating student usage of
  web-based instruction
• An approach that is easily adaptable to different
  types of courses, different population sizes, and
  different attributes to be analyzed
• Rigorous application of known classifiers as a
  means of analyzing and comparing use and
  performance of students who have taken a
  technical course that was partially/completely
  administered via the web

23
Clustering

• Statement of problem
• Classification (Prediction student performance)
• Clustering (Ensembles of multiple clusterings)
• B. Minaei-Bidgoli, A. Topchy and W.F. Punch,
  Ensembles of Partitions via Data Resampling,
  Proc. Intl. Conf. on Information Technology,
  ITCC/IEEE 2004, in press
• B. Minaei-Bidgoli, A. Topchy and W.F. Punch,
  Effect of the Resampling Methods on Clustering
  Ensemble Efficacy, prepared to submit to Intl.
  Conf. on Machine Learning Models, Technologies
  and Applications, 2004
• A. Topchy, B. Minaei-Bigoli, A.K. Jain, W.F.
  Punch, Adaptive Clustering Ensembles, submitted
  to Intl. Conf on Pattern Recognition, ICPR 2004
• Proposed work
• Tentative Schedule

24
Motivation

• Combinations of classifiers proved to be very
  effective in supervised learning framework, e.g.
  bagging and boosting algorithms
• In LON-CAPA, the course and student data are
  distributed
• Distributed data mining requires efficient
  algorithms capable to integrate the solutions
  obtained from multiple sources of data and
  features
• Ensembles of clusterings can provide novel,
  robust, and stable solutions

25
Taxonomy of Clustering Combination Approaches
26
Resampling Methods

• Bootstrapping (Sampling with replacement)
• Create an artificial list by randomly drawing N
  elements from that list. Some elements will be
  picked more than once.
• Statistically on average 37 of elements are
  repeated
• Subsampling (Sampling without replacement)
• Control over the size of subsample

27
Related work on bootstrap partitioning

• Estimate the number of clusters
• (Jain Moreau1987), (Fridlyand Dudoit 2001),
• Clustering validity/reliability
• (Jain Moreau1987), (Fischer Buhmann 2003)
• Find a measure for clustering stability
• (Ben-Hur et. al, 2002),
• Clustering combination
• (Fridlyand Dudoit 2001) (Fischer Buhmann 2003)
• (Monti, et al., 2003.)

28
Experiment Data sets
29
Two-spiral and Halfrings data sets
Halfrings 400 patterns (100-300)
2-Spirals 200 patterns (100-100)
30
Bootstrap results on Iris
31
Subsampling on Halfrings
32
Subsampling results on Galaxy/Star
33
Error Rate for Individual Clustering
34
Summary of the best results of Bootstrap
35
(No Transcript)
36
Additional Proposed work

• Statement of problem
• Classification (Prediction student performance)
• Clustering (Ensembles of multiple clusterings)
• Additional Proposed work
• Association Analysis
• Dynamic mining
• Tentative Schedule

37
A sequence-based clustering

• The problem
• given students browsing data and course
  contents, find clusters of learners with similar
  behavior
• order of browsed pages matters
• P ? P ? R1? R2 ? P ? A
• R1? R2 ? P ? A
• P ? A
• P ? P ? P ? P ? P ? P ? P ? P ? P ? A
• P ? P ? P ? P ? P
• R3 ? R2 ? P
• Cluster students based on a similarity function

38
Web usage mining

• Many techniques have been investigated in the
  e-commerce and CRM
• Some can be adapted, some can not
• The goals are different,
• The user model is different
• Analyzing students interactions with the
  LON-CAPA and take actions accordingly. It is the
  path traversal pattern or similar to web
  sequential pattern mining or web log mining.

39
Association Analysis postprocessing

• Association rules mining studies the frequency of
  items occurring together in a given set of data.
• Solving the discretization problem for continuous
  features
• Post analysis of the discovered knowledge in
  terms of the interestingness, usefulness and so
  on. What is useful or interesting is a domain
  dependent, need to talk to LON-CAPA
  instructors/authors
• Strategic use of data and discovered knowledge

40
Dynamic mining LON-CAPA Examples
You are about to start a test. Other students
similar to you, who succeeded in this test, have
also accessed Section 5 of Chapter 3. You did
not. Would you like to access it now before
attempting the test? Yes No
Based on your time access to solve the problem
(Circular Motion), It seems that you are not
thinking about the problem, It is better to see
the following pages and then submit your
answers Motion in 2 Dimensions Force and Motion
Momentum and Collisions
Someone answered the question you posted on the
Bulletin Board yesterday. Would you like to read
it now? Yes No
Degree of difficulty of problem 3 in homework
set 5 jumped into greater than 90 in the first
5 hours of student access? There might be
something wrong in designing the problem. Would
you like to revise it now? Yes
No
41
Conclusion Enhancing web-based learning

• L-C servers are tracking students activities in
  large logs
• The knowledge discovered could be analyzed and
  evaluated by knowledge experts (off-line mining)
• Integrated mining The patterns discovered are
  fed back to a system that seamlessly and
  transparently would make systems behave
  intelligently.
• We could pass the data through a magic box to
  find some obscure patterns.
• Tools to recommend tasks, automatically adapt
  course materials
• Tools can be personalized, manually or
  automatically

42
Tentative Schedule