Project
·
Two students
will be a team for each project. Each project will be presented in the class at
the end of the quarter.
·
Project
submission and presentation schedule is in the syllabus.
·
Alternative
project: Students may work on
another project with other tasks and competition data with instructor’s
approval. Other Data mining competitions can be found on the web. If
this is a current competition you can submit your forecast to organizers.
Part 1
·
Find on the web two packages of
data mining software (freeware and commercial)
·
Learn the capabilities of
that software
·
Fill table below for each
software similar to done for Tanagra in the table.
·
Write a word file Part1.docs
·
Post your report
in the directory "Project" on your class account (U drive).
|
Software, web link,
type (e.g., freeware, commercial)
|
List of functional capabilities (e.g., data entry,
editing, classification, clustering)
|
Names of the classes of algorithms implemented
(e.g., Decision tree, Neural Networks, k-Nearest Neighbors, Linear
Discrimination)
|
brief description of the ideas of one
of the implemented algorithms (1-4
sentences)
|
Software and Hardware Platform(E.g., windows, PC)
|
|
1.Tanagra,
freeware
|
exploratory data analysis, statistical learning, machine
learning and databases, supervised learning algorithms, the data management.
an architecture allowing users to easily add own data
mining methods, to compare their performances.
free access to source code for learning DM programming
techniques. Stream diagram.
Data source Visualization Descriptive statistics Instance selection Feature selection Feature construction
|
an
interactive and visual construction of decision trees;
Regression Factorial analysis Clustering Supervised learning Meta-spv learning Learning assessment Association
Binary
logistic regression,
K-Nearest
neighbor,
Multi-layer
perceptron,
Decision
Trees,
Naive
Bayes,
Radial
basis function, Prototype-NN
|
Decision tree
implementation is ID3 algorithm with the following steps <describe
steps>.
ID3 :
J.R. Quinlan, "Discovering rules by induction from large collections of
examples", D. Michie ed., Expert Systems in
the Microelectronic age, pp. 168-201, 1979.
|
Windows,
PC.
|
|
2.
|
|
|
|
|
|
3.
|
|
|
|
|
Tips: Check data mining software from major
industry companies such as IBM, Microsoft, Silicon Graphics, SAS, SPSS and
Oracle.
Useful links to start
Data
Mining Software Tools (review)
Tools
for learning Computational Intelligence (download)
Part 2
· Write a summary of the task using the sources
below
o Winner’s paper: C. Elkan:
Magical Thinking in
Data Mining: Lessons From CoIL Challenge 2000
(postscript) ( pdf). Proceedings of the Seventh International
Conference on Knowledge Discovery and Data Mining (KDD'01).
· Build a model and a forecast using data
located at \neve\cs456\data\competition.
· Post your report including the task description
as a word files Project2.docs to your Project2 directory.
Details
1. Data (see Insurance
company problem. File neve\cs456\data\COIL
competition\Analysis\analysispositive.xls contains some preliminary data
analysis.)
1.1. Use the first half of data as training (file neve\cs456\data\COIL competition\COIL-Data\training data
5822 with target & header\ticdata2000h.txt).
1.2. Use the second half of these data as testing data.
1.3. Use file ticeval2000ht.txt (neve\cs456\data\COIL
competition\COIL-Data\validation data with target and header\ ticeval2000ht.txt) for forecasting
the target for each record in this file. Generate forecast as a single
forecasting column, see instructions in
CoIL
Challenge 2000.
2.
Building Data Mining Model
Develop a data
mining model by using Tanagra or other software. See references about
successful algorithms for this task in CoIL
Challenge 2000.
Write a word file Pafrt2.docs as a
report with
Accuracy of the model for training data (include
confusion matrix)
Accuracy of the model for testing data (include confusion matrix)
Comparison of forecast for 1.3 with the result of the CoIL
winner
Description of the model (the complete model as a file model.txt should
be in your directory too)
Post your report as a word file Part2.docs
on your class U drive.
Present your
results in the class
Part 3
1.
Develop your data mining model for data from CWU
AL data
(data location neve\cs456\data\cwudata\).
2. Make a presentation.
3. Write a report on your forecast
with a brief description of the model as an executive summary for CWU AL
association
Later it can be submitted to the
foundation based on peer evaluation and instructor's judgments.
·
Develop data mining model by
using Tanagra or other data mining software
·
Write down a report with
model accuracy for training data
model accuracy for testing
data.
comparison of forecast for 1.3 with at least with one
student in the class
description of the model (the complete model as a file model.txt
should be in your directory)
·
Post your report (make link
to your model.txt file and other related info)
·
Present your model to the
class.
Evaluation
metrics
|
Criterion
|
|
Completeness of the report
|
|
Accuracy on training (% of correct forecast, compare with
our own)
|
|
Accuracy on testing (% of correct forecast, compare with
our own)
|
|
Quality of explanation of the model (the user should be
convinced that the model reasonably explains the forecast)
|
|
Complexity of the model
|
|
Clarity and readability of the Report
|
|
Quality of presentation
|