Classification and Prediction

• What is classification? What is • Support Vector Machines (SVM)

prediction? • Associative classification
• Issues regarding classification • Lazy learners (or learning from
and prediction your neighbors)

• Classification by decision tree • Other classification methods

induction • Prediction

• Bayesian classification • Accuracy and error measures

• Rule-based classification • Ensemble methods

• Model selection
• Classification by back
propagation • Summary

1
 Classification vs. Prediction

• Classification
– predicts categorical class labels (discrete or nominal)
– classifies data (constructs a model) based on the training
set and the values (class labels) in a classifying attribute
and uses it in classifying new data
• Prediction
– models continuous-valued functions, i.e., predicts unknown
or missing values
• Typical applications
– Credit approval
– Target marketing
– Medical diagnosis
– Fraud detection

2
 Classification—A Two-Step Process
• Model construction: describing a set of predetermined classes
– Each tuple/sample is assumed to belong to a predefined class, as
determined by the class label attribute
– The set of tuples used for model construction is training set
– The model is represented as classification rules, decision trees, or
mathematical formulae
• Model usage: for classifying future or unknown objects
– Estimate accuracy of the model
• The known label of test sample is compared with the classified
result from the model
• Accuracy rate is the percentage of test set samples that are
correctly classified by the model
• Test set is independent of training set, otherwise over-fitting will
occur
– If the accuracy is acceptable, use the model to classify data tuples
whose class labels are not known
3
 Process (1): Model Construction

Classification
Algorithms
Training
Data

NAME RANK YEARS TENURED Classifier

M ike A ssistant P rof 3 no (Model)
M ary A ssistant P rof 7 yes
B ill P rofessor 2 yes
Jim A ssociate P rof 7 yes
IF rank = ‘professor’
D ave A ssistant P rof 6 no
OR years > 6
A nne A ssociate P rof 3 no
THEN tenured = ‘yes’
4
Process (2): Using the Model in Prediction

Classifier

Testing
Data Unseen Data

(Jeff, Professor, 4)
NAME RANK YEARS TENURED
T om A ssistant P rof 2 no Tenured?
M erlisa A ssociate P rof 7 no
G eorge P rofessor 5 yes
Joseph A ssistant P rof 7 yes
5
 Supervised vs. Unsupervised Learning

• Supervised learning (classification)

– Supervision: The training data (observations,
measurements, etc.) are accompanied by labels indicating
the class of the observations
– New data is classified based on the training set
• Unsupervised learning (clustering)
– The class labels of training data is unknown
– Given a set of measurements, observations, etc. with the
aim of establishing the existence of classes or clusters in
the data

6
 Issues: Data Preparation

• Data cleaning
– Preprocess data in order to reduce noise and handle
missing values
• Relevance analysis (feature selection)
– Remove the irrelevant or redundant attributes
• Data transformation
– Generalize and/or normalize data

7
 Issues: Evaluating Classification Methods

• Accuracy
– classifier accuracy: predicting class label
– predictor accuracy: guessing value of predicted attributes
• Speed
– time to construct the model (training time)
– time to use the model (classification/prediction time)
• Robustness: handling noise and missing values
• Scalability: efficiency in disk-resident databases
• Interpretability
– understanding and insight provided by the model
• Other measures, e.g., goodness of rules, such as decision tree
size or compactness of classification rules

8
 Decision Tree Induction: Training Dataset

age income student credit_rating buys_computer

<=30 high no fair no
This follows <=30 high no excellent no
31…40 high no fair yes
an >40 medium no fair yes
example of >40 low yes fair yes
Quinlan’s >40 low yes excellent no
31…40 low yes excellent yes
ID3 (Playing <=30 medium no fair no
Tennis) <=30 low yes fair yes
>40 medium yes fair yes
<=30 medium yes excellent yes
31…40 medium no excellent yes
31…40 high yes fair yes
>40 medium no excellent no
9
 Output: A Decision Tree for “buys_computer”

age?

<=30 overcast
31..40 >40

student? yes credit rating?

no yes excellent fair

no yes yes

10
 Algorithm for Decision Tree Induction

• Basic algorithm (a greedy algorithm)

– Tree is constructed in a top-down recursive divide-and-conquer manner
– At start, all the training examples are at the root
– Attributes are categorical (if continuous-valued, they are discretized in
advance)
– Examples are partitioned recursively based on selected attributes
– Test attributes are selected on the basis of a heuristic or statistical
measure (e.g., information gain)
• Conditions for stopping partitioning
– All samples for a given node belong to the same class
– There are no remaining attributes for further partitioning – majority
voting is employed for classifying the leaf
– There are no samples left

11
Attribute Selection Measure: Information
Gain (ID3/C4.5)

 Select the attribute with the highest information gain

 Let pi be the probability that an arbitrary tuple in D belongs to
class Ci, estimated by |Ci, D|/|D|
 Expected information (entropy) needed to classify a tuple in D:
m
Info( D)   pi log 2 ( pi )
i 1

 Information needed (after using A to split D into v partitions) to

classify D: v |D |
InfoA ( D)    I (D j )
j

j 1 | D |
 Information gained by branching on attribute A
Gain(A)  Info(D)  InfoA(D)
12
 Attribute Selection: Information Gain

 Class P: buys_computer = “yes” 5 4

Infoage ( D)  I (2,3)  I (4,0)
 Class N: buys_computer = “no” 14 14
9 9 5 5 5
Info( D)  I (9,5)   log 2 ( )  log 2 ( ) 0.940  I (3,2)  0.694
14 14 14 14 14
age pi ni I(pi, ni) 5
I (2,3)means “age <=30” has 5 out of
<=30 2 3 0.971 14 14 samples, with 2 yes’es and 3
31…40 4 0 0 no’s. Hence
>40 3 2 0.971
age income student credit_rating buys_computer Gain(age)  Info( D)  Infoage ( D)  0.246
<=30 high no fair no
<=30 high no excellent no Similarly,
31…40 high no fair yes
>40 medium no fair yes
>40 low yes fair yes
>40
31…40
low
low
yes excellent
yes excellent
no
yes Gain(income)  0.029
Gain( student )  0.151
<=30 medium no fair no
<=30 low yes fair yes
>40 medium yes fair yes
<=30
31…40
medium
medium
yes excellent
no excellent
yes
yes
Gain(credit _ rating )  0.048
31…40 high yes fair yes
>40 medium no excellent no
13
Computing Information-Gain for Continuous-
Value Attributes
• Let attribute A be a continuous-valued attribute
• Must determine the best split point for A
– Sort the value A in increasing order
– Typically, the midpoint between each pair of adjacent values
is considered as a possible split point
• (ai+ai+1)/2 is the midpoint between the values of ai and ai+1
– The point with the minimum expected information
requirement for A is selected as the split-point for A
• Split:
– D1 is the set of tuples in D satisfying A ≤ split-point, and D2 is
the set of tuples in D satisfying A > split-point
14
 Gain Ratio for Attribute Selection (C4.5)

• Information gain measure is biased towards attributes with a

large number of values
• C4.5 (a successor of ID3) uses gain ratio to overcome the
problem (normalization to information gain)
v | Dj | | Dj |
SplitInfo A ( D)    log 2 ( )
j 1 |D| |D|
– GainRatio(A) = Gain(A)/SplitInfo(A)
4 4 6 6 4 4
• Ex. SplitInfo A ( D)  
14
 log 2 ( )   log 2 ( )   log 2 ( )  0.926
14 14 14 14 14
– gain_ratio(income) = 0.029/0.926 = 0.031
• The attribute with the maximum gain ratio is selected as the
splitting attribute
15
 Gini index (CART, IBM IntelligentMiner)

• If a data set D contains examples from n classes, gini index, gini(D) is defined
as n
gini( D) 1  p 2j
j 1
where pj is the relative frequency of class j in D
• If a data set D is split on A into two subsets D1 and D2, the gini index gini(D) is
defined as
|D1| |D |
gini A (D)  gini(D1)  2 gini(D2)
• Reduction in Impurity: |D| |D|

gini( A)  gini(D)  giniA(D)

• The attribute provides the smallest ginisplit(D) (or the largest reduction in
impurity) is chosen to split the node (need to enumerate all the possible
splitting points for each attribute)
16
 Gini index (CART, IBM IntelligentMiner)

• Ex. D has 9 tuples in buys_computer = “yes” and 5 in “no”

2 2
9 5
gini ( D)  1        0.459
 14   14 
• Suppose the attribute income partitions D into 10 in D1: {low, medium} and 4
in D2  10  4
gini ( D)   Gini ( D )   Gini ( D )
income{low, medium} 1 1
 14   14 

but gini{medium,high} is 0.30 and thus the best since it is the lowest
• All attributes are assumed continuous-valued
• May need other tools, e.g., clustering, to get the possible split values
• Can be modified for categorical attributes

17
 Comparing Attribute Selection Measures

• The three measures, in general, return good results but

– Information gain:
• biased towards multivalued attributes
– Gain ratio:
• tends to prefer unbalanced splits in which one partition is
much smaller than the others
– Gini index:
• biased to multivalued attributes
• has difficulty when # of classes is large
• tends to favor tests that result in equal-sized partitions
and purity in both partitions
18
 Other Attribute Selection Measures

• CHAID: a popular decision tree algorithm, measure based on χ2 test for

independence
• C-SEP: performs better than info. gain and gini index in certain cases
• G-statistics: has a close approximation to χ2 distribution
• MDL (Minimal Description Length) principle (i.e., the simplest solution is
preferred):
– The best tree as the one that requires the fewest # of bits to both (1)
encode the tree, and (2) encode the exceptions to the tree
• Multivariate splits (partition based on multiple variable combinations)
– CART: finds multivariate splits based on a linear comb. of attrs.
• Which attribute selection measure is the best?
– Most give good results, none is significantly superior than others

19
 Overfitting and Tree Pruning

• Overfitting: An induced tree may overfit the training data

– Too many branches, some may reflect anomalies due to noise or outliers
– Poor accuracy for unseen samples
• Two approaches to avoid overfitting
– Prepruning: Halt tree construction early—do not split a node if this
would result in the goodness measure falling below a threshold
• Difficult to choose an appropriate threshold
– Postpruning: Remove branches from a “fully grown” tree—get a
sequence of progressively pruned trees
• Use a set of data different from the training data to decide which is
the “best pruned tree”

20
 Enhancements to Basic Decision Tree Induction

• Allow for continuous-valued attributes

– Dynamically define new discrete-valued attributes that
partition the continuous attribute value into a discrete set of
intervals
• Handle missing attribute values
– Assign the most common value of the attribute
– Assign probability to each of the possible values
• Attribute construction
– Create new attributes based on existing ones that are
sparsely represented
– This reduces fragmentation, repetition, and replication
21
 Classification in Large Databases

• Classification—a classical problem extensively studied by

statisticians and machine learning researchers
• Scalability: Classifying data sets with millions of examples and
hundreds of attributes with reasonable speed
• Why decision tree induction in data mining?
– relatively faster learning speed (than other classification
methods)
– convertible to simple and easy to understand classification
rules
– can use SQL queries for accessing databases
– comparable classification accuracy with other methods

22
 Scalable Decision Tree Induction Methods

• SLIQ (EDBT’96 — Mehta et al.)

– Builds an index for each attribute and only class list and the
current attribute list reside in memory
• SPRINT (VLDB’96 — J. Shafer et al.)
– Constructs an attribute list data structure
• PUBLIC (VLDB’98 — Rastogi & Shim)
– Integrates tree splitting and tree pruning: stop growing the
tree earlier
• RainForest (VLDB’98 — Gehrke, Ramakrishnan & Ganti)
– Builds an AVC-list (attribute, value, class label)
• BOAT (PODS’99 — Gehrke, Ganti, Ramakrishnan & Loh)
– Uses bootstrapping to create several small samples

23
 Scalability Framework for RainForest

• Separates the scalability aspects from the criteria that

determine the quality of the tree
• Builds an AVC-list: AVC (Attribute, Value, Class_label)
• AVC-set (of an attribute X )
– Projection of training dataset onto the attribute X and
class label where counts of individual class label are
aggregated
• AVC-group (of a node n )
– Set of AVC-sets of all predictor attributes at the node n

24
 Rainforest: Training Set and Its AVC Sets

Training Examples AVC-set on Age AVC-set on income

age income studentcredit_rating
buys_computerAge Buy_Computer income Buy_Computer

<=30 high no fair no yes no

<=30 high no excellent no yes no
high 2 2
31…40 high no fair yes <=30 3 2
31..40 4 0 medium 4 2
>40 medium no fair yes
>40 low yes fair yes >40 3 2 low 3 1
>40 low yes excellent no
31…40 low yes excellent yes
AVC-set on
<=30 medium no fair no AVC-set on Student
credit_rating
<=30 low yes fair yes
student Buy_Computer
>40 medium yes fair yes Credit
Buy_Computer

<=30 medium yes excellent yes yes no rating yes no

31…40 medium no excellent yes yes 6 1 fair 6 2
31…40 high yes fair yes no 3 4 excellent 3 3
>40 medium no excellent no
25
 Data Cube-Based Decision-Tree Induction

• Integration of generalization with decision-tree induction

(Kamber et al.’97)
• Classification at primitive concept levels
– E.g., precise temperature, humidity, outlook, etc.
– Low-level concepts, scattered classes, bushy classification-
trees
– Semantic interpretation problems
• Cube-based multi-level classification
– Relevance analysis at multi-levels
– Information-gain analysis with dimension + level
26
 BOAT (Bootstrapped Optimistic Algorithm for Tree
Construction)

• Use a statistical technique called bootstrapping to create

several smaller samples (subsets), each fits in memory
• Each subset is used to create a tree, resulting in several
trees
• These trees are examined and used to construct a new
tree T’
– It turns out that T’ is very close to the tree that would
be generated using the whole data set together
• Adv: requires only two scans of DB, an incremental alg.

27
Presentation of Classification Results

28
Visualization of a Decision Tree in SGI/MineSet 3.0

29
Interactive Visual Mining by Perception-Based
Classification (PBC)

30
 Bayesian Classification: Why?

• A statistical classifier: performs probabilistic prediction, i.e.,

predicts class membership probabilities
• Foundation: Based on Bayes’ Theorem.
• Performance: A simple Bayesian classifier, naïve Bayesian
classifier, has comparable performance with decision tree and
selected neural network classifiers
• Incremental: Each training example can incrementally
increase/decrease the probability that a hypothesis is correct —
prior knowledge can be combined with observed data
• Standard: Even when Bayesian methods are computationally
intractable, they can provide a standard of optimal decision
making against which other methods can be measured

31
 Bayesian Theorem: Basics

• Let X be a data sample (“evidence”): class label is unknown

• Let H be a hypothesis that X belongs to class C
• Classification is to determine P(H|X), the probability that the
hypothesis holds given the observed data sample X
• P(H) (prior probability), the initial probability
– E.g., X will buy computer, regardless of age, income, …
• P(X): probability that sample data is observed
• P(X|H) (posteriori probability), the probability of observing the
sample X, given that the hypothesis holds
– E.g., Given that X will buy computer, the prob. that X is 31..40,
medium income
32
 Bayesian Theorem

• Given training data X, posteriori probability of a hypothesis H,

P(H|X), follows the Bayes theorem

P(H | X)  P(X | H )P(H )

P(X)
• Informally, this can be written as
posteriori = likelihood x prior/evidence
• Predicts X belongs to C2 iff the probability P(Ci|X) is the highest
among all the P(Ck|X) for all the k classes
• Practical difficulty: require initial knowledge of many
probabilities, significant computational cost
33
 Towards Naïve Bayesian Classifier

• Let D be a training set of tuples and their associated class labels,

and each tuple is represented by an n-D attribute vector X = (x1,
x2, …, xn)
• Suppose there are m classes C1, C2, …, Cm.
• Classification is to derive the maximum posteriori, i.e., the
maximal P(Ci|X)
• This can be derived from Bayes’ theorem
P(X | C )P(C )
P(C | X)  i i
i P(X)
• Since P(X) is constant for all classes, only
P(C | X)  P(X | C )P(C )
i i i
needs to be maximized

34
 Derivation of Naïve Bayes Classifier

• A simplified assumption: attributes are conditionally

independent (i.e., no dependence relation between attributes):
n
P( X | C i )   P( x | C i )  P( x | C i )  P( x | C i )  ...  P( x | C i )
k 1 2 n
k 1
• This greatly reduces the computation cost: Only counts the
class distribution
• If Ak is categorical, P(xk|Ci) is the # of tuples in Ci having value xk
for Ak divided by |Ci, D| (# of tuples of Ci in D)
• If Ak is continous-valued, P(xk|Ci) is usually computed based on
Gaussian distribution with a mean μ and standard deviation σ
( x )2
1 
g ( x,  ,  )  e 2 2

and P(xk|Ci) is 2 

P(X | C i)  g ( xk , Ci ,  Ci )
35
 Naïve Bayesian Classifier: Training Dataset
age income studentcredit_rating
buys_compu
<=30 high no fair no
<=30 high no excellent no
Class: 31…40 high no fair yes
C1:buys_computer = ‘yes’ >40 medium no fair yes
C2:buys_computer = ‘no’ >40 low yes fair yes
>40 low yes excellent no
Data sample
31…40 low yes excellent yes
X = (age <=30,
Income = medium, <=30 medium no fair no
Student = yes <=30 low yes fair yes
Credit_rating = Fair) >40 medium yes fair yes
<=30 medium yes excellent yes
31…40 medium no excellent yes
31…40 high yes fair yes
>40 medium no excellent no
36
 Naïve Bayesian Classifier: An Example

• P(Ci): P(buys_computer = “yes”) = 9/14 = 0.643

P(buys_computer = “no”) = 5/14= 0.357

• Compute P(X|Ci) for each class

P(age = “<=30” | buys_computer = “yes”) = 2/9 = 0.222
P(age = “<= 30” | buys_computer = “no”) = 3/5 = 0.6
P(income = “medium” | buys_computer = “yes”) = 4/9 = 0.444
P(income = “medium” | buys_computer = “no”) = 2/5 = 0.4
P(student = “yes” | buys_computer = “yes) = 6/9 = 0.667
P(student = “yes” | buys_computer = “no”) = 1/5 = 0.2
P(credit_rating = “fair” | buys_computer = “yes”) = 6/9 = 0.667
P(credit_rating = “fair” | buys_computer = “no”) = 2/5 = 0.4

• X = (age <= 30 , income = medium, student = yes, credit_rating = fair)

P(X|Ci) : P(X|buys_computer = “yes”) = 0.222 x 0.444 x 0.667 x 0.667 = 0.044

P(X|buys_computer = “no”) = 0.6 x 0.4 x 0.2 x 0.4 = 0.019
P(X|Ci)*P(Ci) : P(X|buys_computer = “yes”) * P(buys_computer = “yes”) = 0.028
P(X|buys_computer = “no”) * P(buys_computer = “no”) = 0.007

Therefore, X belongs to class (“buys_computer = yes”)

37
 Avoiding the 0-Probability Problem

• Naïve Bayesian prediction requires each conditional prob. be non-zero.

Otherwise, the predicted prob. will be zero
n
P( X | C i )   P( x k | C i)
k 1

• Ex. Suppose a dataset with 1000 tuples, income=low (0), income= medium
(990), and income = high (10),
• Use Laplacian correction (or Laplacian estimator)
– Adding 1 to each case
Prob(income = low) = 1/1003
Prob(income = medium) = 991/1003
Prob(income = high) = 11/1003
– The “corrected” prob. estimates are close to their “uncorrected”
counterparts
38
 Naïve Bayesian Classifier: Comments

• Advantages
– Easy to implement
– Good results obtained in most of the cases
• Disadvantages
– Assumption: class conditional independence, therefore loss of
accuracy
– Practically, dependencies exist among variables
• E.g., hospitals: patients: Profile: age, family history, etc.
Symptoms: fever, cough etc., Disease: lung cancer, diabetes, etc.
• Dependencies among these cannot be modeled by Naïve Bayesian
Classifier
• How to deal with these dependencies?
– Bayesian Belief Networks

39
 Bayesian Belief Networks

• Bayesian belief network allows a subset of the variables

conditionally independent
• A graphical model of causal relationships
– Represents dependency among the variables
– Gives a specification of joint probability distribution
 Nodes: random variables
 Links: dependency
X Y  X and Y are the parents of Z, and Y is the
parent of P
Z  No dependency between Z and P
P
 Has no loops or cycles
40
 Bayesian Belief Network: An Example

Family The conditional probability table (CPT)

Smoker
History for variable LungCancer:
(FH, S) (FH, ~S) (~FH, S) (~FH, ~S)

LC 0.8 0.5 0.7 0.1

LungCancer Emphysema ~LC 0.2 0.5 0.3 0.9

CPT shows the conditional probability for each

possible combination of its parents

PositiveXRay Dyspnea Derivation of the probability of a

particular combination of values of X,
from CPT:
n
Bayesian Belief Networks P ( x1 ,..., xn )   P ( x i | Parents(Y i ))
i 1
41
 Training Bayesian Networks

• Several scenarios:
– Given both the network structure and all variables
observable: learn only the CPTs
– Network structure known, some hidden variables: gradient
descent (greedy hill-climbing) method, analogous to neural
network learning
– Network structure unknown, all variables observable:
search through the model space to reconstruct network
topology
– Unknown structure, all hidden variables: No good
algorithms known for this purpose
• Ref. D. Heckerman: Bayesian networks for data mining

42
 Using IF-THEN Rules for Classification
• Represent the knowledge in the form of IF-THEN rules
R: IF age = youth AND student = yes THEN buys_computer = yes
– Rule antecedent/precondition vs. rule consequent
• Assessment of a rule: coverage and accuracy
– ncovers = # of tuples covered by R
– ncorrect = # of tuples correctly classified by R
coverage(R) = ncovers /|D| /* D: training data set */
accuracy(R) = ncorrect / ncovers
• If more than one rule is triggered, need conflict resolution
– Size ordering: assign the highest priority to the triggering rules that has the
“toughest” requirement (i.e., with the most attribute test)
– Class-based ordering: decreasing order of prevalence or misclassification cost per
class
– Rule-based ordering (decision list): rules are organized into one long priority list,
according to some measure of rule quality or by experts

43
 Rule Extraction from a Decision Tree
age
?
<=30 31..40 >40
 Rules are easier to understand than large trees
student? credit rating?
yes
 One rule is created for each path from the root to a
n yes excellent fair
leaf o
no yes yes
 Each attribute-value pair along a path forms a
conjunction: the leaf holds the class prediction
 Rules are mutually exclusive and exhaustive
• Example: Rule extraction from our buys_computer decision-tree
IF age = young AND student = no THEN buys_computer = no
IF age = young AND student = yes THEN buys_computer = yes
IF age = mid-age THEN buys_computer = yes
IF age = old AND credit_rating = excellent THEN buys_computer = yes
IF age = young AND credit_rating = fair THEN buys_computer = no

44
 Rule Extraction from the Training Data

• Sequential covering algorithm: Extracts rules directly from training data

• Typical sequential covering algorithms: FOIL, AQ, CN2, RIPPER
• Rules are learned sequentially, each for a given class Ci will cover many tuples
of Ci but none (or few) of the tuples of other classes
• Steps:
– Rules are learned one at a time
– Each time a rule is learned, the tuples covered by the rules are removed
– The process repeats on the remaining tuples unless termination condition,
e.g., when no more training examples or when the quality of a rule
returned is below a user-specified threshold
• Comp. w. decision-tree induction: learning a set of rules simultaneously

45
 How to Learn-One-Rule?

• Star with the most general rule possible: condition = empty

• Adding new attributes by adopting a greedy depth-first strategy
– Picks the one that most improves the rule quality
• Rule-Quality measures: consider both coverage and accuracy
– Foil-gain (in FOIL & RIPPER): assesses info_gain by extending condition
pos' pos
FOIL _ Gain  pos'(log 2  log 2 )
It favors rules that have high accuracy and ' negmany
poscover ' pos  neg
positive tuples
• Rule pruning based on an independent set of test tuples

pos  neg
FOIL _ Prune( R) 
 neg by R.
poscovered
Pos/neg are # of positive/negative tuples
If FOIL_Prune is higher for the pruned version of R, prune R

46
 Classification: A Mathematical Mapping

• Classification:
– predicts categorical class labels
• E.g., Personal homepage classification
– xi = (x1, x2, x3, …), yi = +1 or –1
– x1 : # of a word “homepage”
– x2 : # of a word “welcome”
• Mathematically
– x  X = n, y  Y = {+1, –1}
– We want a function f: X  Y

47
 Linear Classification

• Binary Classification problem

• The data above the red line
belongs to class ‘x’
• The data below red line
x belongs to class ‘o’
x
x x x • Examples: SVM, Perceptron,
x Probabilistic Classifiers
x x o
x
o
x o o
ooo
o o
o o o o

48
 Discriminative Classifiers

• Advantages
– prediction accuracy is generally high
• As compared to Bayesian methods – in general
– robust, works when training examples contain errors
– fast evaluation of the learned target function
• Bayesian networks are normally slow
• Criticism
– long training time
– difficult to understand the learned function (weights)
• Bayesian networks can be used easily for pattern discovery
– not easy to incorporate domain knowledge
• Easy in the form of priors on the data or distributions
49
 Perceptron & Winnow
• Vector: x, w
x2
• Scalar: x, y, w
Input: {(x1, y1), …}
Output: classification function f(x)
f(xi) > 0 for yi = +1
f(xi) < 0 for yi = -1
f(x) => wx + b = 0
or w1x1+w2x2+b = 0

• Perceptron: update W
additively
• Winnow: update W
multiplicatively
x1

50
 SVM—Support Vector Machines

• A new classification method for both linear and nonlinear data

• It uses a nonlinear mapping to transform the original training
data into a higher dimension
• With the new dimension, it searches for the linear optimal
separating hyperplane (i.e., “decision boundary”)
• With an appropriate nonlinear mapping to a sufficiently high
dimension, data from two classes can always be separated by a
hyperplane
• SVM finds this hyperplane using support vectors (“essential”
training tuples) and margins (defined by the support vectors)

51
 SVM—History and Applications

• Vapnik and colleagues (1992)—groundwork from Vapnik &

Chervonenkis’ statistical learning theory in 1960s
• Features: training can be slow but accuracy is high owing to
their ability to model complex nonlinear decision boundaries
(margin maximization)
• Used both for classification and prediction
• Applications:
– handwritten digit recognition, object recognition, speaker
identification, benchmarking time-series prediction tests

52
 SVM—General Philosophy

Small Margin Large Margin

Support Vectors

53
SVM—Margins and Support
Vectors

54
 SVM—When Data Is Linearly Separable

Let data D be (X1, y1), …, (X|D|, y|D|), where Xi is the set of training tuples associated
with the class labels yi
There are infinite lines (hyperplanes) separating the two classes but we want to find
the best one (the one that minimizes classification error on unseen data)
SVM searches for the hyperplane with the largest margin, i.e., maximum marginal
hyperplane (MMH)

55
 SVM—Linearly Separable
 A separating hyperplane can be written as
W●X+b=0
where W={w1, w2, …, wn} is a weight vector and b a scalar (bias)
 For 2-D it can be written as
w0 + w1 x1 + w2 x2 = 0
 The hyperplane defining the sides of the margin:
H1: w0 + w1 x1 + w2 x2 ≥ 1 for yi = +1, and
H2: w0 + w1 x1 + w2 x2 ≤ – 1 for yi = –1
 Any training tuples that fall on hyperplanes H1 or H2 (i.e., the
sides defining the margin) are support vectors
 This becomes a constrained (convex) quadratic optimization problem:
Quadratic objective function and linear constraints  Quadratic
Programming (QP)  Lagrangian multipliers

56
Why Is SVM Effective on High Dimensional Data?

 The complexity of trained classifier is characterized by the # of support

vectors rather than the dimensionality of the data
 The support vectors are the essential or critical training examples —they lie
closest to the decision boundary (MMH)
 If all other training examples are removed and the training is repeated, the
same separating hyperplane would be found
 The number of support vectors found can be used to compute an (upper)
bound on the expected error rate of the SVM classifier, which is independent
of the data dimensionality
 Thus, an SVM with a small number of support vectors can have good
generalization, even when the dimensionality of the data is high

57
 A2

SVM—Linearly Inseparable

A1
 Transform the original input data into a higher dimensional
space

 Search for a linear separating hyperplane in the new space

58
 SVM—Kernel functions
 Instead of computing the dot product on the transformed data tuples, it is
mathematically equivalent to instead applying a kernel function K(Xi, Xj) to
the original data, i.e., K(Xi, Xj) = Φ(Xi) Φ(Xj)
 Typical Kernel Functions

 SVM can also be used for classifying multiple (> 2) classes and for regression
analysis (with additional user parameters)

59
 Scaling SVM by Hierarchical Micro-Clustering

• SVM is not scalable to the number of data objects in terms of training time
and memory usage
• “Classifying Large Datasets Using SVMs with Hierarchical Clusters Problem”
by Hwanjo Yu, Jiong Yang, Jiawei Han, KDD’03
• CB-SVM (Clustering-Based SVM)
– Given limited amount of system resources (e.g., memory), maximize the
SVM performance in terms of accuracy and the training speed
– Use micro-clustering to effectively reduce the number of points to be
considered
– At deriving support vectors, de-cluster micro-clusters near “candidate
vector” to ensure high classification accuracy

60
 CB-SVM: Clustering-Based SVM

• Training data sets may not even fit in memory

• Read the data set once (minimizing disk access)
– Construct a statistical summary of the data (i.e., hierarchical clusters)
given a limited amount of memory
– The statistical summary maximizes the benefit of learning SVM
• The summary plays a role in indexing SVMs
• Essence of Micro-clustering (Hierarchical indexing structure)
– Use micro-cluster hierarchical indexing structure
• provide finer samples closer to the boundary and coarser samples
farther from the boundary
– Selective de-clustering to ensure high accuracy
61
CF-Tree: Hierarchical Micro-cluster

62
 CB-SVM Algorithm: Outline

• Construct two CF-trees from positive and negative data sets

independently
– Need one scan of the data set
• Train an SVM from the centroids of the root entries
• De-cluster the entries near the boundary into the next level
– The children entries de-clustered from the parent entries
are accumulated into the training set with the non-
declustered parent entries
• Train an SVM again from the centroids of the entries in the
training set
• Repeat until nothing is accumulated

63
 Selective Declustering

• CF tree is a suitable base structure for selective declustering

• De-cluster only the cluster Ei such that
– Di – Ri < Ds, where Di is the distance from the boundary to the
center point of Ei and Ri is the radius of Ei
– Decluster only the cluster whose subclusters have possibilities to
be the support cluster of the boundary
• “Support cluster”: The cluster whose centroid is a support
vector

64
Experiment on Synthetic Dataset

65
Experiment on a Large Data Set

66
 SVM vs. Neural Network

• SVM • Neural Network

– Relatively new concept – Relatively old
– Deterministic algorithm – Nondeterministic algorithm
– Generalizes well but
– Nice Generalization doesn’t have strong
properties mathematical foundation
– Hard to learn – learned in – Can easily be learned in
batch mode using incremental fashion
quadratic programming – To learn complex
functions—use multilayer
techniques
perceptron (not that trivial)
– Using kernels can learn
very complex functions
67
 What Is Prediction?

• (Numerical) prediction is similar to classification

– construct a model
– use model to predict continuous or ordered value for a given input
• Prediction is different from classification
– Classification refers to predict categorical class label
– Prediction models continuous-valued functions
• Major method for prediction: regression
– model the relationship between one or more independent or predictor
variables and a dependent or response variable
• Regression analysis
– Linear and multiple regression
– Non-linear regression
– Other regression methods: generalized linear model, Poisson regression,
log-linear models, regression trees
68
 Linear Regression

• Linear regression: involves a response variable y and a single predictor

variable x
y = w0 + w1 x
where w0 (y-intercept) and w1 (slope) are regression coefficients
• Method of least squares: estimates the best-fitting straight line
| D|

 (x  x )( yi  y )
w  w  y w x
i
i 1

1 | D|
0 1
 (x
i 1
i  x )2

• Multiple linear regression: involves more than one predictor variable

– Training data is of the form (X1, y1), (X2, y2),…, (X|D|, y|D|)
– Ex. For 2-D data, we may have: y = w0 + w1 x1+ w2 x2
– Solvable by extension of least square method or using SAS, S-Plus
– Many nonlinear functions can be transformed into the above
69
 Nonlinear Regression

• Some nonlinear models can be modeled by a polynomial function

• A polynomial regression model can be transformed into linear
regression model. For example,
y = w0 + w1 x + w2 x2 + w3 x3
convertible to linear with new variables: x2 = x2, x3= x3
y = w0 + w1 x + w2 x2 + w3 x3
• Other functions, such as power function, can also be transformed
to linear model
• Some models are intractable nonlinear (e.g., sum of exponential
terms)
– possible to obtain least square estimates through extensive
calculation on more complex formulae
70
 Other Regression-Based Models

• Generalized linear model:

– Foundation on which linear regression can be applied to modeling
categorical response variables
– Variance of y is a function of the mean value of y, not a constant
– Logistic regression: models the prob. of some event occurring as a linear
function of a set of predictor variables
– Poisson regression: models the data that exhibit a Poisson distribution
• Log-linear models: (for categorical data)
– Approximate discrete multidimensional prob. distributions
– Also useful for data compression and smoothing
• Regression trees and model trees
– Trees to predict continuous values rather than class labels

71
 Regression Trees and Model Trees

• Regression tree: proposed in CART system (Breiman et al. 1984)

– CART: Classification And Regression Trees
– Each leaf stores a continuous-valued prediction
– It is the average value of the predicted attribute for the training tuples
that reach the leaf
• Model tree: proposed by Quinlan (1992)
– Each leaf holds a regression model—a multivariate linear equation for
the predicted attribute
– A more general case than regression tree
• Regression and model trees tend to be more accurate than linear regression
when the data are not represented well by a simple linear model

72
 Predictive Modeling in Multidimensional Databases

• Predictive modeling: Predict data values or construct

generalized linear models based on the database data
• One can only predict value ranges or category distributions
• Method outline:
– Minimal generalization
– Attribute relevance analysis
– Generalized linear model construction
– Prediction
• Determine the major factors which influence the prediction
– Data relevance analysis: uncertainty measurement, entropy
analysis, expert judgement, etc.
• Multi-level prediction: drill-down and roll-up analysis

73
Prediction: Numerical Data

74
Prediction: Categorical Data

75
 C1 C2

Classifier Accuracy Measures C1 True positive False negative

C2 False positive True negative

classes buy_computer = yes buy_computer = no total recognition(%)

buy_computer = yes 6954 46 7000 99.34
buy_computer = no 412 2588 3000 86.27
total 7366 2634 10000 95.52
• Accuracy of a classifier M, acc(M): percentage of test set tuples that are
correctly classified by the model M
– Error rate (misclassification rate) of M = 1 – acc(M)
– Given m classes, CMi,j, an entry in a confusion matrix, indicates # of tuples in
class i that are labeled by the classifier as class j
• Alternative accuracy measures (e.g., for cancer diagnosis)
sensitivity = t-pos/pos /* true positive recognition rate */
specificity = t-neg/neg /* true negative recognition rate */
precision = t-pos/(t-pos + f-pos)
accuracy = sensitivity * pos/(pos + neg) + specificity * neg/(pos + neg)
– This model can also be used for cost-benefit analysis

76
 Predictor Error Measures

• Measure predictor accuracy: measure how far off the predicted value is from
the actual known value
• Loss function: measures the error betw. yi and the predicted value yi’
– Absolute error: | yi – yi’|
– Squared error: (yi – yi’)2
• Test error (generalization error):
d
the average loss over the test setd
– Mean absolute error:  | y  yMean
i 1
'| squared error:
i i ( y  y ')
i 1
i i
2

d d
d

– Relative absolute error:

d

 | y Relative
y '|
i squared error:
i
 ( yi  yi ' ) 2
i 1
i 1
d d
| y
i 1
i y|
(y
i 1
i  y)2
The mean squared-error exaggerates the presence of outliers
Popularly use (square) root mean-square error, similarly, root relative
squared error
77
 Evaluating the Accuracy of a Classifier or Predictor (I)
• Holdout method
– Given data is randomly partitioned into two independent sets
• Training set (e.g., 2/3) for model construction
• Test set (e.g., 1/3) for accuracy estimation
– Random sampling: a variation of holdout
• Repeat holdout k times, accuracy = avg. of the accuracies obtained
• Cross-validation (k-fold, where k = 10 is most popular)
– Randomly partition the data into k mutually exclusive subsets, each
approximately equal size
– At i-th iteration, use Di as test set and others as training set
– Leave-one-out: k folds where k = # of tuples, for small sized data
– Stratified cross-validation: folds are stratified so that class dist. in each
fold is approx. the same as that in the initial data

78
 Evaluating the Accuracy of a Classifier or Predictor (II)

• Bootstrap
– Works well with small data sets
– Samples the given training tuples uniformly with replacement
• i.e., each time a tuple is selected, it is equally likely to be selected
again and re-added to the training set
• Several boostrap methods, and a common one is .632 boostrap
– Suppose we are given a data set of d tuples. The data set is sampled d times, with
replacement, resulting in a training set of d samples. The data tuples that did not
make it into the training set end up forming the test set. About 63.2% of the
original data will end up in the bootstrap, and the remaining 36.8% will form the
test set (since (1 – 1/d)d ≈ e-1 = 0.368)
– Repeat the sampling procedue k times, overall accuracy of the model:
k
acc( M )   (0.632  acc( M i )test _ set 0.368  acc( M i )train_ set )
i 1

79
 Ensemble Methods: Increasing the Accuracy

• Ensemble methods
– Use a combination of models to increase accuracy
– Combine a series of k learned models, M1, M2, …, Mk, with
the aim of creating an improved model M*
• Popular ensemble methods
– Bagging: averaging the prediction over a collection of
classifiers
– Boosting: weighted vote with a collection of classifiers
– Ensemble: combining a set of heterogeneous classifiers
80
 Bagging: Boostrap Aggregation

• Analogy: Diagnosis based on multiple doctors’ majority vote

• Training
– Given a set D of d tuples, at each iteration i, a training set Di of d tuples is
sampled with replacement from D (i.e., boostrap)
– A classifier model Mi is learned for each training set Di
• Classification: classify an unknown sample X
– Each classifier Mi returns its class prediction
– The bagged classifier M* counts the votes and assigns the class with the
most votes to X
• Prediction: can be applied to the prediction of continuous values by taking
the average value of each prediction for a given test tuple
• Accuracy
– Often significant better than a single classifier derived from D
– For noise data: not considerably worse, more robust
– Proved improved accuracy in prediction
81
 Boosting

• Analogy: Consult several doctors, based on a combination of weighted

diagnoses—weight assigned based on the previous diagnosis accuracy
• How boosting works?
– Weights are assigned to each training tuple
– A series of k classifiers is iteratively learned
– After a classifier Mi is learned, the weights are updated to allow the
subsequent classifier, Mi+1, to pay more attention to the training tuples
that were misclassified by Mi
– The final M* combines the votes of each individual classifier, where the
weight of each classifier's vote is a function of its accuracy
• The boosting algorithm can be extended for the prediction of continuous
values
• Comparing with bagging: boosting tends to achieve greater accuracy, but it
also risks overfitting the model to misclassified data
82
 Adaboost (Freund and Schapire, 1997)

• Given a set of d class-labeled tuples, (X1, y1), …, (Xd, yd)

• Initially, all the weights of tuples are set the same (1/d)
• Generate k classifiers in k rounds. At round i,
– Tuples from D are sampled (with replacement) to form a training set Di
of the same size
– Each tuple’s chance of being selected is based on its weight
– A classification model Mi is derived from Di
– Its error rate is calculated using Di as a test set
– If a tuple is misclssified, its weight is increased, o.w. it is decreased
• Error rate: err(Xj) is the misclassification error of tuple Xj. Classifier Mi
error rate is the sum of the weights of the misclassified tuples:
d
error ( M i )   w j  err ( X j )
• The weight of classifier Mi’s
j
vote is 1  error ( M i )
log
error ( M i )
83
 Model Selection: ROC Curves

• ROC (Receiver Operating Characteristics)

curves: for visual comparison of classification
models
• Originated from signal detection theory
• Shows the trade-off between the true positive
rate and the false positive rate
 Vertical axis represents
• The area under the ROC curve is a measure of
the true positive rate
the accuracy of the model
 Horizontal axis rep. the
• Rank the test tuples in decreasing order: the false positive rate
one that is most likely to belong to the positive  The plot also shows a
class appears at the top of the list diagonal line
• The closer to the diagonal line (i.e., the closer  A model with perfect
the area is to 0.5), the less accurate is the accuracy will have an area
model of 1.0
84