decision tree classification prof. navneet goyal bits, pilani bits c464 – machine learning
TRANSCRIPT
Decision Tree Classification
Prof. Navneet GoyalBITS, Pilani
BITS C464 – Machine Learning
General Approach
Apply
Model
Induction
Deduction
Learn
Model
Model
Tid Attrib1 Attrib2 Attrib3 Class
1 Yes Large 125K No
2 No Medium 100K No
3 No Small 70K No
4 Yes Medium 120K No
5 No Large 95K Yes
6 No Medium 60K No
7 Yes Large 220K No
8 No Small 85K Yes
9 No Medium 75K No
10 No Small 90K Yes 10
Tid Attrib1 Attrib2 Attrib3 Class
11 No Small 55K ?
12 Yes Medium 80K ?
13 Yes Large 110K ?
14 No Small 95K ?
15 No Large 67K ? 10
Test Set
Learningalgorithm
Training Set
Figure taken from text book (Tan, Steinbach, Kumar)
Decision tree – is a classification scheme
Represents – a model of different classes Generates – tree & set of rules
A node without children - is a leaf node. Otherwise an internal node. Each internal node has - an associated splitting predicate. e.g. binary predicates. Example predicates:
Age <= 20Profession in {student, teacher}
5000*Age + 3*Salary – 10000 > 0
Classification by Decision Tree Induction
Classification by Decision Tree Induction
Decision tree A flow-chart-like tree structure Internal node denotes a test on an attribute Branch represents an outcome of the test Leaf nodes represent class labels or class distribution
Decision tree generation consists of two phases Tree construction
At start, all the training examples are at the root Partition examples recursively based on selected attributes
Tree pruning Identify and remove branches that reflect noise or outliers
Use of decision tree: Classifying an unknown sample Test the attribute values of the sample against the decision tree
Decision tree classifiers are very popular. WHY?
It does not require any domain knowledge or parameter setting, and is therefore suitable for exploratory knowledge discoveryDTs can handle high dimensional dataRepresentation of acquired knowledge in tree form is intuitive and easy to assimilate by humansLearning and classification steps are simple & fastGood accuracy
Classification by Decision Tree Induction
Main AlgorithmsHunt’s algorithmID3C4.5CARTSLIQ,SPRINT
Classification by Decision Tree Induction
Example of a Decision Tree
Tid Refund MaritalStatus
TaxableIncome Cheat
1 Yes Single 125K No
2 No Married 100K No
3 No Single 70K No
4 Yes Married 120K No
5 No Divorced 95K Yes
6 No Married 60K No
7 Yes Divorced 220K No
8 No Single 85K Yes
9 No Married 75K No
10 No Single 90K Yes10
categ
orical
categ
orical
contin
uous
class
Training Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Splitting Attributes
Model: Decision Tree
Figure taken from text book (Tan, Steinbach, Kumar)
Another Example of Decision Tree
Tid Refund MaritalStatus
TaxableIncome Cheat
1 Yes Single 125K No
2 No Married 100K No
3 No Single 70K No
4 Yes Married 120K No
5 No Divorced 95K Yes
6 No Married 60K No
7 Yes Divorced 220K No
8 No Single 85K Yes
9 No Married 75K No
10 No Single 90K Yes10
categ
orical
categ
orical
contin
uous
class
MarSt
Refund
TaxInc
YESNO
NO
NO
Yes No
Married Single,
Divorced
< 80K > 80K
There could be more than one tree that fits the same data!
Figure taken from text book (Tan, Steinbach, Kumar)
Which tree is better and why? How many decision trees? How to find the optimal tree? Is it computationally feasible?(Try constructing a suboptimal tree in reasonable amount of time – greedy algorithm) What should be the order of split? Look for answers in “20 questions” & “Guess Who” games!
Some Questions
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test DataStart from the root of tree.
Figure taken from text book (Tan, Steinbach, Kumar)
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test Data
Figure taken from text book (Tan, Steinbach, Kumar)
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test Data
Figure taken from text book (Tan, Steinbach, Kumar)
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test Data
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test Data
Figure taken from text book (Tan, Steinbach, Kumar)
Apply Model to Test Data
Refund
MarSt
TaxInc
YESNO
NO
NO
Yes No
Married Single, Divorced
< 80K > 80K
Refund Marital Status
Taxable Income Cheat
No Married 80K ? 10
Test Data
Assign Cheat to “No”
Figure taken from text book (Tan, Steinbach, Kumar)
Decision Trees: Example Training Data Set
No PlayTrue6068Rain
No PlayFalse7066Rain
PlayFalse6078Rain
PlayFalse9588Overcast
PlayTrue7563Overcast
PlayFalse8888Overcast
No PlayTrue9060Sunny
PlayTrue7579Sunny
PlayFalse7056Sunny
No playtrue9079Sunny
ClassWindyHumidityTempOutlook
Numerical Attributes
Temprature, Humidity
Categorical Attributes
Outlook, Windy
Class ???
Class label
Sample Decision Tree
Outlook
sunny rain
Humidity
true false<=75
Play No
> 75
{1}
Play Windy
PlayNo PlayNo Play
overcast
Five leaf nodes – Each represents a rule
Decision Trees: Example
Decision Trees: Example
Rules corresponding to the given tree
1. If it is a sunny day and humidity is not above 75%, then play.
2. If it is a sunny day and humidity is above 75%, then do not play.
3. If it is overcast, then play.
4. If it is rainy and not windy, then play.
5. If it is rainy and windy, then do not play.
Is it the best classification ????
Decision Trees: Example
Accuracy of the classifier
determined by the percentage of the test data set that is correctly classified
Class: “No Play”
Classification of new record
New record: outlook=rain, temp =70, humidity=65,
windy=true.
Decision Trees: Example
Test Data Set
PlayTrue6068Rain
No PlayFalse7066Rain
PlayFalse6078Rain
PlayFalse9588Overcast
PlayTrue7563Overcast
No PlayFalse8888Overcast
No PlayTrue9060Sunny
No PlayTrue7579Sunny
PlayFalse7056Sunny
Playtrue9079Sunny
ClassWindyHumidityTempOutlook Rule 1: two records
Sunny & hum <=75
(one is correctly classified)Accuracy= 50%
Rule 2:sunny, hum> 75
Accuracy = 50%Rule 3: overcastAccuracy= 66%
Practical Issues of Classification
Underfitting and OverfittingMissing ValuesCosts of Classification
Overfitting the Data
A classification model commits two kinds of errors: Training Errors (TE) (resubstitution, apparent errors) Generalization Errors (GE)
A good classification model must have low TE as well as low GE
A model that fits the training data too well can have high GE than a model with high TE
This problem is known as model overfitting
Underfitting and OverfittingOverfitting
Underfitting: when model is too simple, both training and test errors are large. TE & GE are large when the size of the tree is very small.
It occurs because the model is yet to learn the true structure of the data and as a result it performs poorly on both training and test sets
Figure taken from text book (Tan, Steinbach, Kumar)
Overfitting the Data
When a decision tree is built, many of the branches may reflect anomalies in the training data due to noise or outliers.
We may grow the tree just deeply enough to perfectly classify the training data set.
This problem is known as overfitting the data.
Overfitting the Data
TE of a model can be reduced by increasing the model complexity
Leaf nodes of the tree can be expanded until it perfectly fits the training data
TE for such a complex tree = 0GE can be large because the tree may
accidently fit noise points in the training setOverfitting & underfitting are two pathologies
that are related to model complexity
Occam’s RazorGiven two models of similar generalization
errors, one should prefer the simpler model over the more complex model
For complex models, there is a greater chance that it was fitted accidentally by errors in data
Therefore, one should include model complexity when evaluating a model
Definition
A decision tree T is said to overfit the training data if there exists some other tree T’ which is a simplification of T, such that T has smaller error than T’ over the training set but T’ has a smaller error than T over the entire distribution of the instances.
Problems of Overfitting
Overfitting can lead to many difficulties: Overfitted models are incorrect. Require more space and more computational
resources Require collection of unnecessary features They are more difficult to comprehend
Overfitting
Overfitting can be due to:
1. Presence of Noise
2. Lack of representative samples
Overfitting: Example
Presence of Noise: Training SetName Body
TemperatureGives Birth
4-legged Hibernates Class Label(mammal)
Procupine Warm Blooded Y Y Y Y
Cat Warm Blooded Y Y N Y
Bat Warm Blooded Y N Y N*
Whale Warm Blooded Y N N N*
Salamander Cold Blooded N Y Y N
Komodo dragon Cold Blooded N Y N N
Python Cold Blooded N N Y N
Salmon Cold Blooded N N N N
Eagle Warm Blooded N N N N
Guppy Cold Blooded Y N N N
Table taken from text book (Tan, Steinbach, Kumar)
Overfitting: Example
Presence of Noise: Training SetName Body
TemperatureGives Birth
4-legged Hibernates Class Label(mammal)
Procupine Warm Blooded Y Y Y Y
Cat Warm Blooded Y Y N Y
Bat Warm Blooded Y N Y N*
Whale Warm Blooded Y N N N*
Salamander Cold Blooded N Y Y N
Komodo dragon Cold Blooded N Y N N
Python Cold Blooded N N Y N
Salmon Cold Blooded N N N N
Eagle Warm Blooded N N N N
Guppy Cold Blooded Y N N N
Table taken from text book (Tan, Steinbach, Kumar)
Overfitting: Example
Presence of Noise:Test SetName Body
TemperatureGives Birth
4-legged Hibernates Class Label(mammal)
Human Warm Blooded Y N N Y
Pigeon Warm Blooded N N N N
Elephant Warm Blooded Y Y N Y
Leopard Shark Cold Blooded Y N N N
Turtle Cold Blooded N Y N N
Penguin Cold Blooded N N N N
Eel Cold Blooded N N N N
Dolphin Warm Blooded Y N N Y
Spiny Anteater Warm Blooded N Y Y Y
Gila Monster Cold Blooded N Y Y N
Table taken from text book (Tan, Steinbach, Kumar)
Overfitting: Example
Presence of Noise: Models
Body Temp
4-legged
Mammals Non-mammals
Yes
Warm blooded
Gives BirthNoYes
No
Non-mammals
Non-mammals
Cold bloodedBody Temp
Mammals Non-mammals
Warm blooded
Gives BirthNoYes
Cold blooded
Non-mammals
Model M1TE = 0%, GE=30% Find out why?
Model M2TE = 20%, GE=10%
Figure taken from text book (Tan, Steinbach, Kumar)
Overfitting: Example
Lack of representative samples: Training Set
Name Body Temperature
Gives Birth
4-legged Hibernates Class Label(mammal)
Salamander Cold Blooded N Y Y N
Eagle Warm Blooded N N N N
Guppy Cold Blooded Y N N N
Poorwill Warm blooded N N Y N
Platypus Warm blooded N Y Y Y
Table taken from text book (Tan, Steinbach, Kumar)
Overfitting: Example
Lack of representative samples: Training Set
Body Temp
4-legged
Mammals Non-mammals
Yes
Warm blooded
HibernatesNoYes
No
Non-mammals
Non-mammals
Cold blooded
Model M3TE = 0%, GE=30% Find out why?
Figure taken from text book (Tan, Steinbach, Kumar)
Overfitting due to Noise
Decision boundary is distorted by noise point
Figure taken from text book (Tan, Steinbach, Kumar)
Overfitting due to Insufficient Examples
Lack of data points in the lower half of the diagram makes it difficult to predict correctly the class labels of that region
- Insufficient number of training records in the region causes the decision tree to predict the test examples using other training records that are irrelevant to the classification task
Figure taken from text book (Tan, Steinbach, Kumar)
How to Address Overfitting
Pre-Pruning (Early Stopping Rule) Stop the algorithm before it becomes a fully-grown tree Typical stopping conditions for a node:
Stop if all instances belong to the same class Stop if all the attribute values are the same
More restrictive conditions: Stop if number of instances is less than some user-specified
threshold Stop if class distribution of instances are independent of the
available features (e.g., using 2 test) Stop if expanding the current node does not improve impurity
measures (e.g., Gini or information gain).
How to Address Overfitting…
Post-pruningGrow decision tree to its entiretyTrim the nodes of the decision tree in a
bottom-up fashionIf generalization error improves after
trimming, replace sub-tree by a leaf node.Class label of leaf node is determined
from majority class of instances in the sub-tree
Can use MDL for post-pruning
Post-pruning
Post-pruning approach- removes branches of a fully grown tree. Subtree replacement replaces a subtree with a
single leaf node
Alt
Price
Yes Yes No
$$$
$$$
Yes
AltYes
Yes
Post-pruning
Subtree raising moves a subtree to a higher level in the decision tree, subsuming its parent
Alt
Price
Yes Yes No
$$$
$$$
Yes
ResYesNo
No4/4
Alt
Price
Yes Yes No
$$$
$$$
Yes
Overfitting: Example
Presence of Noise:Training SetName Body
TemperatureGives Birth
4-legged Hibernates Class Label(mammal)
Porcupine Warm Blooded Y Y Y Y
Cat Warm Blooded Y Y N Y
Bat Warm Blooded Y N Y N*
Whale Warm Blooded Y N N N*
Salamander Cold Blooded N Y Y N
Komodo Dragon Cold Blooded N Y N N
Python Cold Blooded N N Y N
Salmon Cold Blooded N N N N
Eagle Warm Blooded N N N N
Guppy Cold Blooded Y N N N
Table taken from text book (Tan, Steinbach, Kumar)
Post-pruning: Techniques
Cost Complexity pruning Algorithm: pruning operation is performed if it does not increase the estimated error rate. Of course, error on the training data is not the useful estimator (would
result in almost no pruning)
Minimum Description Length Algorithm: states that the best tree is the one that can be encoded using the fewest number of bits. The challenge for the pruning phase is to find the subtree that can be
encoded with the least number of bits.
Hunt’s Algorithm Let Dt be the set of training records
that reach a node t Let y={y1,y2,…yc} be the class labels
Step 1: If Dt contains records that belong
the same class yt, then t is a leaf node labeled as yt. If Dt is an empty set, then t is a leaf node labeled by the default class, yd
Step 2: If Dt contains records that belong to
more than one class, use an attribute test to split the data into smaller subsets. Recursively apply the procedure to each child node
Tid Refund Marital Status
Taxable Income Cheat
1 Yes Single 125K No
2 No Married 100K No
3 No Single 70K No
4 Yes Married 120K No
5 No Divorced 95K Yes
6 No Married 60K No
7 Yes Divorced 220K No
8 No Single 85K Yes
9 No Married 75K No
10 No Single 90K Yes 10
Dt
?
Figure taken from text book (Tan, Steinbach, Kumar)
Hunt’s AlgorithmDon’t Cheat
Refund
Don’t Cheat
Don’t Cheat
Yes No
Refund
Don’t Cheat
Yes No
MaritalStatus
Don’t Cheat
Cheat
Single,Divorced Married
TaxableIncome
Don’t Cheat
< 80K >= 80K
Refund
Don’t Cheat
Yes No
MaritalStatus
Don’t CheatCheat
Single,Divorced Married
Figure taken from text book (Tan, Steinbach, Kumar)
Should handle the following additional conditions: Child nodes created in step 2 are empty. When can this happen?
Declare the node as leaf node (majority class label of the training records of parent node)
In step 2, if all the records associated with Dt have identical attributes (except for the class label), then it is not possible to split these records further. Declare the node as leaf with the same class label as the majority class of training records associated with this node.
Hunt’s Algorithm
Tree Induction
Greedy strategy.Split the records based on an attribute
test that optimizes certain criterion.
IssuesDetermine how to split the records
How to specify the attribute test condition?How to determine the best split?
Determine when to stop splitting
Design Issues of Decision Tree Induction How should the training records be split?
At each recursive step, an attribute test condition must be selected. Algorithm must provide a method for specifying the test condition for diff. attrib. types as well as an objective measure for evaluating the goodness of each test condition
How should the splitting procedure stop?
Stopping condition is needed to terminate the tree-growing process. Stop when:
- all records belong to the same class
- all records have identical values
- both conditions are sufficient to stop any DT induction algo., other criterion can be imposed to terminate the procedure early (do we need to do this? Think of model over-fitting!)
Hunt’s Algorithm
How to determine the Best Split?
OwnCar?
C0: 6C1: 4
C0: 4C1: 6
C0: 1C1: 3
C0: 8C1: 0
C0: 1C1: 7
CarType?
C0: 1C1: 0
C0: 1C1: 0
C0: 0C1: 1
StudentID?
...
Yes No Family
Sports
Luxury c1c10
c20
C0: 0C1: 1
...
c11
Before Splitting: 10 records of class 0,10 records of class 1
Which test condition is the best?
Slide taken from text book slides available at companion website (Tan, Steinbach, Kumar)
Greedy approach: Nodes with homogeneous class
distribution are preferredNeed a measure of node impurity:
C0: 5C1: 5
C0: 9C1: 1
Non-homogeneous,
High degree of impurity
Homogeneous,
Low degree of impurity
How to determine the Best Split?
Slide taken from text book slides available at companion website (Tan, Steinbach, Kumar)
Measures of Node Impurity
- Based on the degree of impurity of child nodes
- Less impurity more skew
- node with class distribution (1,0) has zero impurity, whereas a node with class distribution (0.5, 0.5) has highest impurity
Gini IndexEntropyMisclassification error -
Measures of Node Impurity
Gini Index
Entropy
Misclassification error
j
tjptGINI 2)]|([1)(
j
tjptjptEntropy )|(log)|()(
)|(max1)( tiPtErrori
Comparison among Splitting Criteria
For a 2-class problem:
Figure taken from text book (Tan, Steinbach, Kumar)
Impurity
How to find the best split? Example:
Node N1
C0: 0
C1:6
Node N2
C0: 1
C1:5
Node N3
C0: 3
C1:3
GINI = 0
ENTROPY =0
ERROR = 0
GINI = 0.278
ENTROPY = 0.650
ERROR = 0.167
GINI = 0.5
ENTROPY =1
ERROR = .5
How to find the best split? The 3 measures have similar characteristic curves Despite this, the attribute chosen as the test condition may vary
depending on the choice of the impurity measure Need to normalize these measures! Introducing GAIN,
where I=impurity measure of a given node
N = total no. of records at the parent node
K = no. of attribute values
N(vj) = no. of records associated with the child node vj
I(parent) is same for all test conditions When entropy is used, it is called Information Gain, info The larger the Gain, the better is the split Is it the best measure?
)()(
)(1
j
k
j
j vIN
vNparentI
How to Find the Best Split?
B?
Yes No
Node N3 Node N4
A?
Yes No
Node N1 Node N2
Before Splitting:
C0 N10 C1 N11
C0 N20 C1 N21
C0 N30 C1 N31
C0 N40 C1 N41
C0 N00 C1 N01
M0
M1 M2 M3 M4
M12 M34Gain = M0 – M12 vs M0 – M34
Slide taken from text book slides available at companion website (Tan, Steinbach, Kumar)
Measure of Impurity: GINI Gini Index for a given node t :
(NOTE: p( j | t) is the relative frequency of class j at node t).
Maximum (1 - 1/nc) when records are equally distributed among all classes, implying least interesting information
Minimum (0.0) when all records belong to one class, implying most interesting information
j
tjptGINI 2)]|([1)(
C1 0C2 6
Gini=0.000
C1 2C2 4
Gini=0.444
C1 3C2 3
Gini=0.500
C1 1C2 5
Gini=0.278
Slide taken from text book slides available at companion website (Tan, Steinbach, Kumar)
Examples for computing GINI
C1 0 C2 6
C1 2 C2 4
C1 1 C2 5
P(C1) = 0/6 = 0 P(C2) = 6/6 = 1
Gini = 1 – P(C1)2 – P(C2)2 = 1 – 0 – 1 = 0
j
tjptGINI 2)]|([1)(
P(C1) = 1/6 P(C2) = 5/6
Gini = 1 – (1/6)2 – (5/6)2 = 0.278
P(C1) = 2/6 P(C2) = 4/6
Gini = 1 – (2/6)2 – (4/6)2 = 0.444
Splitting Based on GINI Used in CART, SLIQ, SPRINT. When a node p is split into k partitions (children), the
quality of split is computed as,
where, ni = number of records at child i,
n = number of records at node p.
k
i
isplit iGINI
n
nGINI
1
)(
Binary Attributes: Computing GINI Index
Splits into two partitions Effect of Weighing partitions:
– Larger and Purer Partitions are sought for.
A?
Yes No
Node N1 Node N2
Parent
C1 6
C2 6
Gini = 0.500
N1 N2 C1 4 2
C2 3 3
Gini=0.486
Gini(N1) = 1 – (4/7)2 – (3/7)2 = 0.490
Gini(N2) = 1 – (2/5)2 – (3/5)2 = 0.480
Gini(Children) = 7/12 * 0.490 + 5/12 * 0.480= 0.486
Binary Attributes: Computing GINI Index
Splits into two partitions Effect of Weighing partitions:
– Larger and Purer Partitions are sought for.
B?
Yes No
Node N1 Node N2
Parent
C1 6
C2 6
Gini = 0.500
N1 N2 C1 5 1
C2 2 4
Gini=0.371
Gini(N1) = 1 – (5/7)2 – (2/7)2 = 0.408
Gini(N2) = 1 – (1/5)2 – (4/5)2 = 0.320
Gini(Children) = 7/12 * 0.408 + 5/12 * 0.320= 0.371
Attribute B is preferred over A
Categorical Attributes: Computing Gini Index
For each distinct value, gather counts for each class in the dataset
Use the count matrix to make decisions
CarType{Sports,Luxury}
{Family}
C1 3 1
C2 2 4
Gini 0.400
CarType
{Sports}{Family,Luxury}
C1 2 2
C2 1 5
Gini 0.419
CarType
Family Sports Luxury
C1 1 2 1
C2 4 1 1
Gini 0.393
Multi-way split Two-way split (find best partition of values)
GINI favours multiway splits!!
Continuous Attributes: Computing Gini Index
Use Binary Decisions based on one value
Several Choices for the splitting value Number of possible splitting values
= Number of distinct values Each splitting value has a count matrix
associated with it Class counts in each of the
partitions, A < v and A v Simple method to choose best v
For each v, scan the database to gather count matrix and compute its Gini index
Computationally Inefficient! Repetition of work.
TaxableIncome> 80K?
Yes No
Continuous Attributes: Computing Gini Index...
For efficient computation: for each attribute, Sort the attribute on values Linearly scan these values, each time updating the count matrix and
computing gini index Choose the split position that has the least gini index
Cheat No No No Yes Yes Yes No No No No
Taxable Income
60 70 75 85 90 95 100 120 125 220
55 65 72 80 87 92 97 110 122 172 230
<= > <= > <= > <= > <= > <= > <= > <= > <= > <= > <= >
Yes 0 3 0 3 0 3 0 3 1 2 2 1 3 0 3 0 3 0 3 0 3 0
No 0 7 1 6 2 5 3 4 3 4 3 4 3 4 4 3 5 2 6 1 7 0
Gini 0.420 0.400 0.375 0.343 0.417 0.400 0.300 0.343 0.375 0.400 0.420
Split Positions
Sorted Values
Find the time complexity in terms of # records!
Alternative Splitting Criteria based on INFO
Entropy at a given node t:
(NOTE: p( j | t) is the relative frequency of class j at node t).
Measures homogeneity of a node. Maximum (log nc) when records are equally distributed
among all classes implying least informationMinimum (0.0) when all records belong to one class,
implying most information
Entropy based computations are similar to the GINI index computations
j
tjptjptEntropy )|(log)|()(
Examples for computing Entropy
C1 0 C2 6
C1 2 C2 4
C1 1 C2 5
P(C1) = 0/6 = 0 P(C2) = 6/6 = 1
Entropy = – 0 log 0 – 1 log 1 = – 0 – 0 = 0
P(C1) = 1/6 P(C2) = 5/6
Entropy = – (1/6) log2 (1/6) – (5/6) log2 (1/6) = 0.65
P(C1) = 2/6 P(C2) = 4/6
Entropy = – (2/6) log2 (2/6) – (4/6) log2 (4/6) = 0.92
j
tjptjptEntropy )|(log)|()( 2
Splitting Based on INFO...
Information Gain:
Parent Node, p is split into k partitions;
ni is number of records in partition i
Measures Reduction in Entropy achieved because of the split. Choose the split that achieves most reduction (maximizes GAIN)
Used in ID3 and C4.5 Disadvantage: Tends to prefer splits that result in large
number of partitions, each being small but pure.
k
i
i
splitiEntropy
nn
pEntropyGAIN1
)()(
Gain Ratio:
Parent Node, p is split into k partitions
ni is the number of records in partition i
Adjusts Information Gain by the entropy of the partitioning (SplitINFO). Higher entropy partitioning (large number of small partitions) is penalized!
Used in C4.5 Designed to overcome the disadvantage of Information
Gain
SplitINFO
GAINGainRATIO Split
split
k
i
ii
nn
nn
SplitINFO1
log
Splitting Based on INFO...
Splitting Criteria based on Classification Error
Classification error at a node t :
Measures misclassification error made by a node. Maximum (1 - 1/nc) when records are equally
distributed among all classes, implying least interesting information
Minimum (0.0) when all records belong to one class, implying most interesting information
)|(max1)( tiPtErrori
Examples for Computing Error
C1 0 C2 6
C1 2 C2 4
C1 1 C2 5
P(C1) = 0/6 = 0 P(C2) = 6/6 = 1
Error = 1 – max (0, 1) = 1 – 1 = 0
P(C1) = 1/6 P(C2) = 5/6
Error = 1 – max (1/6, 5/6) = 1 – 5/6 = 1/6
P(C1) = 2/6 P(C2) = 4/6
Error = 1 – max (2/6, 4/6) = 1 – 4/6 = 1/3
)|(max1)( tiPtErrori
Misclassification Error vs GiniA?
Yes No
Node N1 Node N2
Parent
C1 7
C2 3
Gini = 0.42
N1 N2 C1 3 4
C2 0 3
Gini=0.361
Gini(N1) = 1 – (3/3)2 – (0/3)2 = 0
Gini(N2) = 1 – (4/7)2 – (3/7)2 = 0.489
Gini(Children) = 3/10 * 0 + 7/10 * 0.489= 0.342
Gini improves !!
Decision Tree Based Classification
Advantages:Inexpensive to constructExtremely fast at classifying unknown
recordsEasy to interpret for small-sized treesAccuracy is comparable to other
classification techniques for many simple data sets
Example: C4.5Simple depth-first construction.Uses Information GainSorts Continuous Attributes at each
node.Needs entire data to fit in memory.Unsuitable for Large Datasets.
Needs out-of-core sorting.
You can download the software from:http://www.cse.unsw.edu.au/~quinlan/c4.5r8.tar.gz
ExampleWeb robot or crawler Based on access patterns, distinguish between
human user and web robots Web Usage Mining BUILD a MODEL – use web log data Think of some more applications of classification!!
Summary: DT Classifiers Does not require any prior assumptions about prob.
dist. Satisfied by classes Finding optimal DT is NP-complete Construction of DT is fast even for large data sets. Testing is also fast. O(w), w=max. depth of the tree Robust to niose Irrelevant attributes can cause problems. (use
feature selection) Data fragmentation problem (leaf nodes having very
few records) Tree pruning has greater impact on the final tree
than choice of impurity measure
Decision Boundary
y < 0.33?
: 0 : 3
: 4 : 0
y < 0.47?
: 4 : 0
: 0 : 4
x < 0.43?
Yes
Yes
No
No Yes No
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
x
y
• Border line between two neighboring regions of different classes is known as decision boundary
• Decision boundary is parallel to axes because test condition involves a single attribute at-a-time
Oblique Decision Trees
x + y < 1
Class = + Class =
• Test condition may involve multiple attributes
• More expressive representation
• Finding optimal test condition is computationally expensive
Tree ReplicationP
Q R
S 0 1
0 1
Q
S 0
0 1
• Same subtree appears in multiple branches
Split using P redundant?
Remove P in post pruning
Metrics for Performance Evaluation
Focus on the predictive capability of a model Rather than how fast it takes to classify or build models,
scalability, etc.
Confusion Matrix:
TP: predicted to be in YES, and is actually in it FP: predicted to be in YES, but is not actually in it TN: predicted not to be in YES, and is not actually in it FN: predicted not to be in YES, but is actually in it
PREDICTED CLASS
ACTUALCLASS
Class=Yes Class=No
Class=Yes a b
Class=No c d
a: TP (true positive)
b: FN (false negative)
c: FP (false positive)
d: TN (true negative)
Metrics for Performance Evaluation…
Most widely-used metric:
PREDICTED CLASS
ACTUALCLASS
Class=Yes Class=No
Class=Yes a(TP)
b(FN)
Class=No c(FP)
d(TN)
FNFPTNTPTNTP
dcbada
Accuracy
Limitation of Accuracy
Consider a 2-class problemNumber of Class 0 examples = 9990Number of Class 1 examples = 10
If model predicts everything to be class 0, accuracy is 9990/10000 = 99.9 %Accuracy is misleading because model
does not detect any class 1 example
Cost Matrix
PREDICTED CLASS
ACTUALCLASS
C(i|j) Class=Yes Class=No
Class=Yes C(Yes|Yes) C(No|Yes)
Class=No C(Yes|No) C(No|No)
C(i|j): Cost of misclassifying class j example as class i
Computing Cost of ClassificationCost
MatrixPREDICTED CLASS
ACTUALCLASS
C(i|j) + -
+ -1 100
- 1 0
Model M1 PREDICTED CLASS
ACTUALCLASS
+ -
+ 150 40
- 60 250
Model M2 PREDICTED CLASS
ACTUALCLASS
+ -
+ 250 45
- 5 200
Accuracy = 80%
Cost = 3910
Accuracy = 90%
Cost = 4255
Cost-Sensitive Measures
cbaa
prrp
baa
caa
222
(F) measure-F
(r) Recall
(p)Precision
Precision is biased towards C(Yes|Yes) & C(Yes|No) Recall is biased towards C(Yes|Yes) & C(No|Yes) F-measure is biased towards all except C(No|No)
dwcwbwawdwaw
4321
41Accuracy Weighted