Transcript Slide 1

Knowledge-Driven
Business Intelligence
Systems: Part II
Week 11
Dr. Jocelyn San Pedro
School of Information Management &
Systems
Monash University
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
Lecture Outline
 Data Mining Technologies
 Neural Networks
 Genetic Algorithms
 Fuzzy Logic
 Decision Trees
 Data Visualisation
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
2
Learning Objectives
At the end of this lecture, the students will
 Gain some understanding of data mining
technologies (decision trees, neural
networks, genetic algorithms, and fuzzy
logic) that are commonly used in data mining
techniques
 Preview some visualisation tools and gain an
understanding of how they support business
decision making
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
3
Data Mining Technologies
1960s – classical statistical analysis
 Correlation, regression, chi-square, cross-tabulation
1980s – classical statistical analysis augmented
by more powerful set of soft computing
techniques
 neural networks, genetic algorithms, fuzzy logic,
decision trees
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
4
Soft Computing
Emerging discipline that combines
computational methods for dealing with
inexact, approximate reasoning
approaches
 simulating the brain-way of solving problems neural networks
 evolving solutions - genetic algorithms
 dealing with logical ambiguity - fuzzy logic
 representing effect of each event, or decision,
on successive events – decision trees
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
5
Neural Networks
 Attempt to mirror the way human brain works in
recognizing patterns by developing mathematical
structures with the ability to learn (Marakas, 2002)
 Attempt to “learn” patterns from data directly, by sifting
data repeatedly, searching for relationships,
automatically building models, and correcting over and
over again the model’s own mistakes – (Dhar and
Stein, 1997)
 Good at modelling poorly understood problems for
which sufficient data can be collected
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
6
Artificial Neural Nets (ANNs)
simple computer programs that build models
from data by trial and error
“Learning from Experience”
 Present a piece of data to a neural network
 The net predicts an output
 The net compares is guess to the actual correct value
(also presented to the network)
 If ANN guess is right, the net does nothing
 If ANN guess is wrong, net figures out how to adjust
some internal parameters so that it can make better
prediction if it sees similar data again in future
 Over time, the ANN begins to converge on a fairly
accurate model of the process
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
7
Artificial Neural Nets (ANNs)
Network Topology- The number of layers and units in each
layer and a way in which the units are connected
together.
3 basic layers:
The input layer receives the data
1. The internal or hidden layer processes the data.
2. The output layer relays the final result of the net.
Output Layer
Guesses
Hidden Layer
Processing
Input Layer
Data Input
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 8
IMS3001
BUSINESS
SYSTEMS
, 2004
Artificial Neural Nets (ANNs)
Make initial guess based on
current weight settings and
inputs
Calculate error
with associated
output
Determine the amount
and direction of individual
weight adjustment
Training the ANN adjusting neural network
weights. During training the
network analyses the data
you have provided and
changes weights between
network units to reflect
dependencies found in your
data.
Adjust individual
weights according
to calculations
Calculate error/adjust
weights for each node in
hidden layer
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 9
IMS3001
BUSINESS
SYSTEMS
, 2004
Artificial Neural Nets (ANNs)
Testing is a process of estimating quality of the trained
neural network. During this process a part of data that
wasn't used during training is presented to the trained
network case by case. Then forecasting error is
measured on each case and used as the estimation of
network quality.
Preparing the ANN in Alyuda Forecaster – www.alyuda.com
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
10
Artificial Neural Nets (ANNs)
 Effective in problems of image recognition
 Not suited well for, say, financial or serious medical
applications.
 highly intricate systems - include dozens of
neurons with a couple hundred connections
between them
 non-transparency of forecasting models
represented by a trained neural network
 knowledge reflected in terms of weights of a couple
hundred intraneural connections cannot be
analysed and interpreted by a human.
 Despite of these difficulties neural networks are
actively used (with varying success) in different
financial applications in the majority of developed
countries.
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
11
ANN Applications – Alyuda
Forecaster
 Credit Approval - determine risk of granting a loan to
an applicant
 Classify applicant as either LOW risk, HIGH risk
 Guide decision in granting or denying new loans
 Employee retention- identify potential employees who
are likely to stay with the organization during the next
year based on previous year data
 Classify employee’s retention probability as LOW or
HIGH probability
 Identify employees who intend to leave and take the
appropriate measures to retain them.
www.alyuda.com
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
12
ANN Applications – Alyuda
Forecaster
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
13
ANN Applications – Alyuda
Forecaster
 Gas consumption - forecast gas consumption by a
power plant.
 Sales forecasting - forecast weekly sales of a small
restaurant chain using the historical data over 109
weeks period
 Stock prediction - forecast the percentage of the Close
price change for Chevron Corp 4 days in advance
www.alyuda.com
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
14
Data Mining Technologies
Genetic Algorithms
 Recognise a good solution, spreads some of that
solution’s features into a population of competing
solutions, and “breeds” good solutions
 Powerful technique for solving various combinatorial or
optimisation problems
 Sample Genetic algorithm online demos
http://math.hws.edu/xJava/GA/
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
15
Genetic Algorithm
 First a population of possible solutions to a problem
are developed.
 Next, the better solutions are recombined with each
other to form some new solutions.
 Finally the new solutions are used to replace the
poorer of the original solutions and the process is
repeated.
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
16
Genetic Algorithm - Example
Selecting a fixed number of market parameters
influencing the market performance the most
 names of these parameters comprise a descriptive set
or a set of chromosomes determining qualities of an
"organism" - a solution of the problem
 Values of parameters determining a solution
correspond to genes
 A search for the optimal solution is similar then to the
process of evolution of a population of organisms,
where each organism is represented by a set of its
chromosomes.
http://www.megaputer.com/dm/systems.php3#stat_package
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
17
Genetic Algorithm - Example
The process of evolution of population of organisms is
driven by three mechanisms:
 selection of the strongest – or survival of the fittest
those sets of chromosomes that characterise the most
optimal solutions
 cross-breeding - production of new organisms by
mixing sets of chromosomes of parent sets of
chromosome
 mutations - accidental changes of genes in some
organisms of the population.
 After a number of new generations built with the help of
the described mechanisms one obtains a solution that
cannot be improved any further. This solution is taken
as a final one.
http://www.megaputer.com/dm/systems.php3#stat_package
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
18
Genetic Algorithms- Weak
Points
 The very way of formulating the problem deprives one
of any opportunity to estimate statistical significance of
the obtained solution.
 Second, only a specialist can develop a criterion for
the chromosome selection and formulate the problem
effectively.
 Thus genetic algorithms should be considered at
present more as an instrument for scientific research
rather than as a tool for generic practical data analysis,
for instance, in finance.
http://www.megaputer.com/dm/systems.php3#stat_package
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
19
Fuzzy Logic
 Our language is full of vague and imprecise concepts,
and allows for conveyance of meaning through
semantic approximations
 These approximations are useful to humans, but do not
readily lend themselves to the rule-based reasoning
done on computers.
 Use of fuzzy logic is how computers handle this
ambiguity
 Allows for partial or “fuzzy” description of rules
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 20
IMS3001
BUSINESS
SYSTEMS
, 2004
The Basics of Fuzzy Logic
 In a “crisp” rule, the result is either false (0) or true (1)
and can be stored in a binary fashion.
 In a “fuzzy” rule, the result ranges from 0 (absolutely
false) to 1 (absolutely true), with stops in between.
 absolutely false, slightly false, slightly true,
absolutely true
 slightly similar, similar, very similar
 These operations utilise functions that assign a degree
of “membership” in a set.
 Degree of similarity of current data to historical data
is 0.75
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 21
IMS3001
BUSINESS
SYSTEMS
, 2004
Membership Function
Example
1.00
Degree of
0.50
Tallness
0.00
0
1
2
3
4
5
6
7
8
9
10
Height in Feet
The “Tallness” function takes a person’s height and
converts it to a numerical scale from 0 to 1.
Here the statement “He is Tall” is absolutely false for
heights below 5 feet and absolutely true for heights
above 7 feet
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 22
IMS3001
BUSINESS
SYSTEMS
, 2004
Inferencing using Fuzzy
Rules
Example
“Well if you’ve got a high margin, price sensitive
product, promoting that product via ads,
displays, etc. is likely to have a high impact on
sales volume. If the volume impact is high, it’s
a good candidate for allocation of promotion
dollars.
But you also want to promote products
more heavily when they’re relatively new in
order to increase market awareness and to
establish market share…”
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
23
Inferencing using Fuzzy
Rules
One fuzzy rule: If product is new, then a client
should spend more money promoting it
new-product-rule
Product is NEW
THEN
Promotion should be HIGH
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
24
Inferencing using Fuzzy
Rules
 - Degree of
Membership in the fuzzy
set NEW

1
0.3
0
0
235
365
Days since product was introduced
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
25
Inferencing using Fuzzy
Rules
Promotion expense that is 2%
of sales is absolutely LOW
The
degree of
“Lowness”
of
Promotion
expense
that is
2.9% of
sales is
0.75.
PROMOTION
1
0.75
Low
Medium
High
0
0
3
5
8
Expense as a percentage of sales
15
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
26
Inferencing using Fuzzy
Rules
Price Sensitivity
1
0.4
(ratio of % change in volume per change in price)Price
sensitivity
is 0.4 LOW
or 0.1
Medium
Low
Medium
High
0.1
0
0
Input
1
2
3
4
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
5
Take Max
value or
Fuzzy Set
Union:
Price
sensitivity
is 0.4 LOW
27
Inferencing using Fuzzy
Rules
Other fuzzy rules:
 If product is NEW, then a client should spend MORE
money promoting it
 If the price sensitivity of product is LOW, then
promotion should be LOW
 If the price sensitivity of product is MEDIUM, then
promotion should be MEDIUM
 If the price sensitivity of product is HIGH, then
promotion should be HIGH
Dhar, V. and Stein, R. (1997)
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
28
Fuzzy Systems
Some Advantages
 Great in dealing with qualitative data, as well as object
attribute
 Offers an attractive trade-off between accuracy and
compactness – express relationships in terms of
simple rules
 Not computationally expensive – compared to “crisp”
rule-based systems
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
29
Fuzzy Systems
Some Disadvantages
 Saturation of fuzzy sets – fuzzy sets get so full of
inferences that the consequent fuzzy regions are
overloaded > system loses the information provided by
the fuzzy rules
 Needs domain expertise to setup fuzzy sets
 Only provides approximation to human reasoning
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
30
Notes on Decision Trees
CART – Classification and Regression Trees
 Most common decision tree, statistical analysis data
mining tool
 automatically searches for and finds high performance
classification and prediction
 key elements are a set of rules for:
 splitting each node in a tree;
 deciding when a tree is complete; and
 assigning each terminal node to a class outcome (or
predicted value for regression)
 More info and software demo on http://www.salfordsystems.com/
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
31
Data Visualisation
For any kind of high dimensional data set, displaying
predictive relationships is a challenge.
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
32
http://www.sapdesignguild.org/editions/edition2/info_zoom.asp
Human Visual Perception and
Data Visualisation
 Data visualisation is so powerful because the human
visual cortex converts objects into information so
quickly.
 The next three slides show (1) usage of global private
networks, (2) flow through natural gas pipelines, and
(3) a risk analysis report that permits the user to draw
an interactive yield curve.
 All three use height or shading to add additional
dimensions to the figure.
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall
IMS3001
BUSINESS
SYSTEMS
, 2004
33
Global Private Network Activity
High Activity
Low Activity
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall 34
IMS3001
BUSINESS
SYSTEMS
, 2004
Natural Gas Pipeline Analysis
Note: Height shows total flow through compressor stations.
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall
IMS3001
BUSINESS
SYSTEMS
, 2004
35
An “Enlivened” Risk Analysis
Report
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall
IMS3001
BUSINESS
SYSTEMS
, 2004
36
Telephone Polling Results
Note: On the “live” map, clicking on an area allows the user
to drill down and see results for smaller areas.
Marakas, –G.M.
(2002) INTELLIGENCE
Decision support
systems–inSEM
the 1
21st
Century. 2nd Ed, Prentice Hall
IMS3001
BUSINESS
SYSTEMS
, 2004
37
References
Dhar, V. and Stein, R. (1997) Intelligent decision Support
Methods: the Science of Knowledge Work, Prentice
Hall.
Dhar, V. and Stein, R. (1997) Seven methods for
transforming corporate data into business intelligence.
Marakas, G.M. (2002) Decision support systems in the
21st Century. 2nd Ed, Prentice Hall (or other editions)
Power, D. (2002) Decision Support Systems: Concepts
and Resources for Managers, Quorum Books.
***********
Good Online resource on fuzzy sets and operations
http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/sb
aa/report.fuzzysets.html
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
38
Questions?
[email protected]
School of Information Management and Systems,
Monash University
T1.28, T Block, Caulfield Campus
9903 2735
IMS3001 – BUSINESS INTELLIGENCE SYSTEMS – SEM 1 , 2004
39