Based on slides from Han J., et. al.

2013

1
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

2
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

3
Prediction Problems: Classification vs. Numeric
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
Numeric Prediction
models continuous-valued functions, i.e., predicts unknown
or missing values
Typical applications
Credit/loan approval:
Medical diagnosis: if a tumor is cancerous or benign
Fraud detection: if a transaction is fraudulent
Web page categorization: which category it is

4
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 overfitting)
 If the accuracy is acceptable, use the model to classify new data
 Note: If the test set is used to select models, it is called validation (test) set

5
Process (1): Model Construction

Classification
Algorithms
Training
Data

NAME RANK YEARS TENURED Classifier

Mike Assistant Prof 3 no (Model)
Mary Assistant Prof 7 yes
Bill Professor 2 yes
Jim Associate Prof 7 yes IF rank = ‘professor’
Dave Assistant Prof 6 no
OR years > 6
Anne Associate Prof 3 no
THEN tenured = ‘yes’
6
Process (2): Using the Model in Prediction

Classifier

Testing
Data Unseen Data

(Jeff, Professor, 4)
NAME RANK YEARS TENURED
Tom Assistant Prof 2 no Tenured?
Merlisa Associate Prof 7 no
George Professor 5 yes
Joseph Assistant Prof 7 yes
7
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

8
 age income student credit_rating buys_computer
<=30 high no fair no
 Training data set: Buys_computer <=30 high no excellent no
 The data set follows an example of 31…40 high no fair yes
>40 medium no fair yes
Quinlan’s ID3 (Playing Tennis) >40 low yes fair yes
 Resulting tree: >40 low yes excellent no
31…40 low yes excellent yes
age? <=30 medium no fair no
<=30 low yes fair yes
>40 medium yes fair yes
<=30 medium yes excellent yes
<=30 overcast
31..40 >40 31…40 medium no excellent yes
31…40 high yes fair yes
>40 medium no excellent no

student? yes credit rating?

no yes excellent fair

no yes no yes
9
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
10
Brief Review of Entropy

m=2
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
m a tuple in D:
Info( D)   pi log 2 ( pi )
i 1
 Information needed (after using A to split D into v partitions) to
v | D |
classify D:
Info A ( D ) 
j
Info( D j )
j 1 | D |

 Information gained by branching on attribute A

Gain(A) Info(D)  Info A(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(p i, n i) 5
<=30 2 3 0.971 I (2,3) means “age <=30” has 5 out of
14 14 samples, with 2 yes’es and 3
31…40 4 0 0
>40 3 2 0.971 no’s. Hence
age
<=30
income student credit_rating
high no fair
buys_computer
no
Gain(age) Info( D )  Infoage ( D ) 0.246
<=30 high no excellent no
31…40 high no fair yes Similarly,
>40 medium no fair yes
>40 low yes fair yes
>40 low yes excellent no
31…40 low yes excellent yes Gain(income) 0.029
<=30 medium no fair no
<=30
>40
low
medium
yes
yes
fair
fair
yes
yes
Gain( student ) 0.151
<=30
31…40
medium
medium
yes
no
excellent
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-Valued 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)
Ex.

gain_ratio(income) = 0.029/1.557 = 0.019

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| |D2 |
gini A ( D)  gini( D1)  gini( D 2)
|D| |D|
Reduction in Impurity: 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
Computation of Gini Index
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
giniincome{low,medium} ( D)   Gini( D1 )    Gini( D2 )
 14   14 

Gini{low,high} is 0.458; Gini{medium,high} is 0.450. Thus, split on the

{low,medium} (and {high}) since it has the lowest Gini index
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-statistic: 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 is decision tree induction popular?
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
RainForest (VLDB’98 — Gehrke, Ramakrishnan & Ganti)
Builds an AVC-list (attribute, value, class label)
22
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

23
 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

yes no
<=30 high no excellent no
high 2 2
31…40 high no fair yes <=30 2 3
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 Buy_Computer
>40 medium yes fair yes
Credit
<=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
24
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.

25
Presentation of Classification Results

Data Mining: Concepts and 26

April 20, 2025 Techniques
Visualization of a Decision Tree in SGI/MineSet 3.0

Data Mining: Concepts and 27

April 20, 2025 Techniques
Data Mining: Concepts and 28
Techniques
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

29
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
30
Bayes’ Theorem: Basics
M
 Total probability Theorem: P(B)   P(B | A )P( A )
i
i 1 i

 Bayes’ Theorem: P(H | X) P(X | H )P(H ) P(X | H )P(H ) / P(X)

P(X)
 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), (i.e., posteriori probability): 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) (likelihood): 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
31
Prediction Based on Bayes’ 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 | H )P( H ) / P(X)

P(X)
Informally, this can be viewed as
posteriori = likelihood x prior/evidence
Predicts X belongs to Ci iff the probability P(Ci|X) is the highest
among all the P(Ck|X) for all the k classes
Practical difficulty: It requires initial knowledge of many
probabilities, involving significant computational cost
32
Deriving the Maximum Posteriori
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
33
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
This greatly 
reduces the computation cost: Only counts the class
k 1
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 
and P(xk|Ci) is g ( x,  ,  )  e 2 2
2 

P ( X | C i )  g ( xk ,  Ci ,  Ci )
34
Naïve Bayes Classifier: Training Dataset

age income studentcredit_rating

buys_compute
Class:
<=30 high no fair no
C1:buys_computer = <=30 high no excellent no
‘yes’ 31…40 high no fair yes
>40 medium no fair yes
C2:buys_computer = ‘no’ >40 low yes fair yes
>40 low yes excellent no
Data to be classified: 31…40 low yes excellent yes
<=30 medium no fair no
X = (age <=30, <=30 low yes fair yes
Income = medium, >40 medium yes fair yes
<=30 medium yes excellent yes
Student = yes 31…40 medium no excellent yes
Credit_rating = Fair) 31…40 high yes fair yes
>40 medium no excellent no

35
Naïve Bayes Classifier: An Example age income studentcredit_rating
buys_com
<=30 high no fair no
 P(C ): P(buys_computer = “yes”) = 9/14 = 0.643 <=30 high no excellent no
i 31…40
>40
high
medium
no fair
no fair
yes
yes
P(buys_computer = “no”) = 5/14= 0.357 >40
>40
low
low
yes fair
yes excellent
yes
no

 Compute P(X|C ) for each class 31…40

<=30
low
medium
yes excellent
no fair
yes
no
i <=30 low yes fair yes

P(age = “<=30” | buys_computer = “yes”) = 2/9 = 0.222

>40 medium yes fair yes
<=30 medium yes excellent yes
31…40 medium no excellent yes

P(age = “<= 30” | buys_computer = “no”) = 3/5 = 0.6 31…40

>40
high
medium
yes fair
no excellent
yes
no

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”) 36
Avoiding the Zero-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
37
Naïve Bayes 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 Bayes
Classifier
How to deal with these dependencies? Bayesian Belief Networks
(Chapter 9)

38
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

39
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
40
Using IF-THEN Rules for Classification
If more than one rule are 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 tests)
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

41
Rule Extraction from a Decision Tree
 Rules are easier to understand than large
trees age?
 One rule is created for each path from the <=30 31..40 >40
root to a leaf student? credit rating?
yes
 Each attribute-value pair along a path forms a
no yes excellent fair
conjunction: the leaf holds the class
no yes no yes
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 = no
IF age = old AND credit_rating = fair THEN buys_computer = yes 42
Rule Induction: Sequential Covering Method
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
Repeat the process on the remaining tuples until 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 43
Sequential Covering Algorithm
while (enough target tuples left)
generate a rule
remove positive target tuples satisfying this rule

Examples covered
Examples covered by Rule 2
by Rule 1 Examples covered
by Rule 3

Positive
examples

44
Rule Generation
To generate a rule
while(true)
find the best predicate p
if foil-gain(p) > threshold then add p to current rule
else break

A3=1&&A1=2
A3=1&&A1=2
&&A8=5A3=1

Positive Negative
examples examples

45
How to Learn-One-Rule?
Start 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 )
pos 'neg ' pos  neg
 favors rules that have high accuracy and cover many positive tuples
Rule pruning based on an independent set of test tuples
pos  neg
FOIL _ Prune( R) 
pos  neg
Pos/neg are # of positive/negative tuples covered by R.
If FOIL_Prune is higher for the pruned version of R, prune R

46
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

47
Model Evaluation and Selection
Evaluation metrics: How can we measure accuracy? Other metrics
to consider?
Use validation test set of class-labeled tuples instead of training set
when assessing accuracy
Methods for estimating a classifier’s accuracy:
 Holdout method, random subsampling
 Cross-validation
 Bootstrap
Comparing classifiers:
 Confidence intervals
 Cost-benefit analysis and ROC Curves

48
Classifier Evaluation Metrics: Confusion Matrix
Confusion Matrix:
Actual class\Predicted class C1 ¬ C1
C1 True Positives (TP) False Negatives (FN)
¬ C1 False Positives (FP) True Negatives (TN)

Example of Confusion Matrix:

Actual class\Predicted buy_computer buy_computer Total
class = yes = no
buy_computer = yes 6954 46 7000
buy_computer = no 412 2588 3000
Total 7366 2634 10000
Given m classes, an entry, CMi,j in a confusion matrix indicates
# of tuples in class i that were labeled by the classifier as class j
May have extra rows/columns to provide totals
49
Classifier Evaluation Metrics: Accuracy,
Error Rate, Sensitivity and Specificity
A\P C ¬C  Class Imbalance Problem:
C TP FN P  One class may be rare, e.g.
¬C FP TN N
fraud, or HIV-positive
P’ N’ All
 Significant majority of the

Classifier Accuracy, or negative class and minority of

recognition rate: percentage of test the positive class
set tuples that are correctly  Sensitivity: True Positive

classified recognition rate

Accuracy = (TP + TN)/All 
Sensitivity = TP/P
Error rate: 1 – accuracy, or  Specificity: True Negative

Error rate = (FP + FN)/All recognition rate


Specificity = TN/N
50
Classifier Evaluation Metrics:
Precision and Recall, and F-measures
Precision: exactness – what % of tuples that the classifier
labeled as positive are actually positive

Recall: completeness – what % of positive tuples did the

classifier label as positive?
Perfect score is 1.0
Inverse relationship between precision & recall
F measure (F or F-score): harmonic mean of precision and
1
recall,

F : weighted measure of precision and recall

ß
assigns ß times as much weight to recall as to precision

51
Classifier Evaluation Metrics: Example

Actual Class\Predicted class cancer = yes cancer = no Total Recognition(%)

cancer = yes 90 210 300 30.00 (sensitivity
cancer = no 140 9560 9700 98.56 (specificity)
Total 230 9770 10000 96.40 (accuracy)
 Precision = 90/230 = 39.13% Recall = 90/300 = 30.00%

52
 Evaluating Classifier Accuracy:
Holdout & Cross-Validation Methods
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 D as test set and others as training set
i
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

53
Evaluating Classifier Accuracy: Bootstrap
 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 bootstrap methods, and a common one is .632 boostrap
 A data set with d tuples 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
end up in the bootstrap, and the remaining 36.8% form the test set (since (1
– 1/d)d ≈ e-1 = 0.368)
 Repeat the sampling procedure k times, overall accuracy of the model:

54
Estimating Confidence Intervals:
Classifier Models M1 vs. M2

Suppose we have 2 classifiers, M1 and M2, which one is better?

Use 10-fold cross-validation to obtain and

These mean error rates are just estimates of error on the true
population of future data cases
What if the difference between the 2 error rates is just attributed to
chance?
 Use a test of statistical significance

 Obtain confidence limits for our error estimates

55
Estimating Confidence Intervals:
Null Hypothesis
Perform 10-fold cross-validation
Assume samples follow a t distribution with k–1 degrees of
freedom (here, k=10)
Use t-test (or Student’s t-test)
Null Hypothesis: M1 & M2 are the same
If we can reject null hypothesis, then
 we conclude that the difference between M1 & M2 is statistically

significant
 Chose model with lower error rate

56
Estimating Confidence Intervals: t-test

If only 1 test set available: pairwise comparison

For ith round of 10-fold cross-validation, the same cross
partitioning is used to obtain err(M1)i and err(M2)i
Average over 10 rounds to get
an
t-test computes t-statistic with k-1 d
degrees of freedom:

where
If two test sets available: use non-paired t-test

wher
e
where k1 & k2 are # of cross-validation samples used for M1 & M2, resp.
57
Estimating Confidence Intervals:
Table for t-distribution

Symmetric
Significance level,
e.g., sig = 0.05 or
5% means M1 & M2
are significantly
different for 95% of
population
Confidence limit, z
= sig/2

58
Estimating Confidence Intervals:
Statistical Significance
Are M1 & M2 significantly different?
 Compute t. Select significance level (e.g. sig = 5%)
 Consult table for t-distribution: Find t value corresponding to k-1
degrees of freedom (here, 9)
 t-distribution is symmetric: typically upper % points of
distribution shown → look up value for confidence limit z=sig/2
(here, 0.025)
 If t > z or t < -z, then t value lies in rejection region:
 Reject null hypothesis that mean error rates of M1 & M2 are same
 Conclude: statistically significant difference between M1 & M2
 Otherwise, conclude that any difference is chance

59
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
 The area under the ROC curve is a  Vertical axis
measure of the accuracy of the model represents the true
 Rank the test tuples in decreasing order: positive rate
the one that is most likely to belong to  Horizontal axis rep.
the positive class appears at the top of the false positive rate
the list

 The plot also shows a
The closer to the diagonal line (i.e., the diagonal line
closer the area is to 0.5), the less  A model with perfect
accurate is the model accuracy will have an
area of 1.0
60
 Issues Affecting Model Selection
Accuracy
classifier accuracy: predicting class label
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
61
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors) Section
9.5
 Summary

62
Lazy vs. Eager Learning
Lazy vs. eager learning
Lazy learning (e.g., instance-based learning): Simply stores
training data (or only minor processing) and waits until it is
given a test tuple
Eager learning (the above discussed methods): Given a set of
training tuples, constructs a classification model before
receiving new (e.g., test) data to classify
Lazy: less time in training but more time in predicting
Accuracy
Lazy method effectively uses a richer hypothesis space since it
uses many local linear functions to form an implicit global
approximation to the target function
Eager: must commit to a single hypothesis that covers the
entire instance space
63
Lazy Learner: Instance-Based Methods
Instance-based learning:
Store training examples and delay the processing (“lazy
evaluation”) until a new instance must be classified
Typical approaches
k-nearest neighbor approach
 Instances represented as points in a Euclidean space.
Locally weighted regression
 Constructs local approximation
Case-based reasoning
 Uses symbolic representations and knowledge-based inference

64
The k-Nearest Neighbor Algorithm
All instances correspond to points in the n-D space
The nearest neighbor are defined in terms of Euclidean
distance, dist(X1, X2)
Target function could be discrete- or real- valued
For discrete-valued, k-NN returns the most common value
among the k training examples nearest to xq
Vonoroi diagram: the decision surface induced by 1-NN for
a typical set of training examples

_
_
_ _ .
+
_
. +
xq +
. . .
_
+ . 65
Discussion on the k-NN Algorithm
k-NN for real-valued prediction for a given unknown tuple
Returns the mean values of the k nearest neighbors
Distance-weighted nearest neighbor algorithm
Weight the contribution of each of the k neighbors according
to their distance to the query xq w 1
d ( xq , x )2
 Give greater weight to closer neighbors i
Robust to noisy data by averaging k-nearest neighbors
Curse of dimensionality: distance between neighbors could be
dominated by irrelevant attributes
To overcome it, axes stretch or elimination of the least
relevant attributes
66
  Requires three things
Unknown record – The set of stored records
– Distance Metric to compute
distance between records
– The value of k, the number of
nearest neighbors to retrieve

 To classify an unknown record:

– Compute distance to other
training records
– Identify k nearest neighbors
– Use class labels of nearest
neighbors to determine the class
label of unknown record (e.g., by
taking majority vote)
 X X X

(a) 1-nearest neighbor (b) 2-nearest neighbor (c) 3-nearest neighbor

K-nearest neighbors of a record x are data points

that have the k smallest distance to x
Nearest Neighbor Classification
Compute distance between two points:
 Euclidean distance

d ( p, q )   ( pi
i
 q)
i
2

Determine the class from nearest neighbor list

 take the majority vote of class labels among the k-nearest
neighbors
 Weigh the vote according to distance
 weight factor, w = 1/d2
Nearest Neighbor Classification…
Choosing the value of k:
 If k is too small, sensitive to noise points
 If k is too large, neighborhood may include points from other classes

X
Nearest Neighbor Classification…
Scaling issues
 Attributes may have to be scaled to prevent distance
measures from being dominated by one of the attributes
 Example:
 height of a person may vary from 1.5m to 1.8m
 weight of a person may vary from 90lb to 300lb

 income of a person may vary from $10K to $1M

Nearest Neighbor Classification…

Problem with Euclidean measure:

 High dimensional data
 curse of dimensionality
 Can produce counter-intuitive results

111111111110 100000000000
vs
011111111111 000000000001

d= d=
1.4142 1.4142
 Solution: Normalize the vectors to unit length
Nearest neighbor Classification…
k-NN classifiers are lazy learners
 It does not build models explicitly
 Unlike eager learners such as decision tree induction and
rule-based systems
 Classifying unknown records are relatively expensive
Chapter 8. Classification
Classification: Basic Concepts
Decision Tree Induction
Bayes Classification Methods
Rule-Based Classification
Model Evaluation and Selection
Lazy Learners (or Learning from Your Neighbors)
Summary

74
Summary (I)
Classification is a form of data analysis that extracts models
describing important data classes.
Effective and scalable methods have been developed for decision
tree induction, Naive Bayesian classification, rule-based
classification, and many other classification methods.
Evaluation metrics include: accuracy, sensitivity, specificity,
precision, recall, F measure, and Fß measure.
Stratified k-fold cross-validation is recommended for accuracy
estimation. Bagging and boosting can be used to increase overall
accuracy by learning and combining a series of individual models.

75
Summary (II)
Significance tests and ROC curves are useful for model selection.
There have been numerous comparisons of the different
classification methods; the matter remains a research topic
No single method has been found to be superior over all others
for all data sets
Issues such as accuracy, training time, robustness, scalability, and
interpretability must be considered and can involve trade-offs,
further complicating the quest for an overall superior method

76
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

77
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 set
d

| y  yi ' | (y i  yi ' ) 2

 Mean absolute error: i 1
i
Mean squared error: i 1

d d
d d
| y i  yi ' | (y i  yi ' ) 2
 Relative absolute error:i d1 Relative squared error: i 1
| y  y| d

(y
i
i 1
i  y)2
i 1

The mean squared-error exaggerates the presence of outliers

Popularly use (square) root mean-square error, similarly, root relative squared
error
78