Decision Tree

R. Akerkar
TMRF, Kolhapur, India

R. Akerkar

Introduction

A classification scheme which generates a tree and g a set of rules from given data set. The t f Th set of records available f d d il bl for developing l i classification methods is divided into two disjoint subsets a training set and a test set. g The attributes of the records are categorise into two types:

Attributes hose Attrib tes whose domain is n merical are called n merical numerical numerical attributes. Attributes whose domain is not numerical are called the categorical attributes attributes.
R. Akerkar 2

Introduction

A decision tree is a tree with the following p p g properties: An inner node represents an attribute. An edge represents a test on the attribute of the father node. node A leaf represents one of the classes. Construction of a decision tree Based on the training data Top Down strategy Top-Down

R. Akerkar

Decision Tree
Example

The data set has five attributes. There is a special attribute: the attribute class is the class label. The attributes, temp (temperature) and humidity are numerical attributes Other attributes are categorical, that is, they cannot be ordered. Based on the training data set, we want to find a set of rules to know what values of outlook, temperature, humidity and wind, determine whether or not to play golf.
R. Akerkar 4

Decision Tree
Example

We have five leaf nodes. In a decision tree, each leaf node represents a rule. We have the following rules corresponding to the tree given in Figure.
RULE 1 RULE 2 RULE 3 RULE 4 RULE 5 If it is sunny and the humidity is not above 75% then play 75%, play. If it is sunny and the humidity is above 75%, then do not play. If it is overcast, then play. If it is rainy and not windy, then play. If it is rainy and windy, then don't play. i i d i d th d 't l
R. Akerkar 5

Classification

The classification of an unknown input vector is done by p y traversing the tree from the root node to a leaf node. A record enters the tree at the root node. At the root, a test is applied to determine which child root node the record will encounter next. This process is repeated until the record arrives at a leaf node. d All the records that end up at a given leaf of the tree are y classified in the same way. There is a unique path from the root to each leaf. The path is a rule which is used to classify the records.

R. Akerkar

In our tree, we can carry out the classification for a u o an unknown record as follows. o eco d o o s Let us assume, for the record, that we know the values of the first four attributes (but we do not know the value of class attribute) as outlook= rain; temp = 70; humidity = 65; and windy true. windy= true
R. Akerkar 7

We start from the root node to check the value of the attribute associated at the root node. This attribute is the splitting attribute at this node. For a decision tree, at every node there is an attribute associated with the node called the splitting attribute. In our example, outlook is the splitting attribute at root. Since for the given record, outlook = rain, we move to the rightmost child node of the root. At this node, the splitting attribute is windy and we find that for the record we want classify, windy = true. Hence, we move to the left child node to conclude that the class label Is "no l " l b l I " play".

R. Akerkar

The accuracy of the classifier is determined by the percentage of the test d t t t data set that is correctly classified. t th t i tl l ifi d We can see that for Rule 1 there are two records of the test data set satisfying outlook= sunny and humidity < 75, and only one of these is correctly classified as play. Thus, the accuracy of this rule is 0.5 (or 50%). Similarly, the accuracy of Rule 2 is also 0.5 (or 50%). The accuracy of Rule 3 is 0.66.

RULE 1 If it is sunny and the humidity is not above 75%, then play.

R. Akerkar

Concept of Categorical Attributes

Consider the following training data set. There are three attributes, namely, age, pincode and class. The attribute class is used for class label.
The attribute age is a numeric attribute, whereas pincode is a categorical one. Though th d Th h the domain of pincode i numeric, no ordering can b d fi d i f i d is i d i be defined among pincode values. You cannot derive any useful information if one pin-code is greater than another pincode pincode.
R. Akerkar 10

Figure gives a decision tree for the training data data. The splitting attribute at the root is pincode and the splitting criterion here is pincode = 500 046. Similarly, for the left child node, the p g g ( splitting criterion is age < 48 (the splitting attribute is age). Although the right child node has the same attribute as the splitting attribute, the splitting criterion is different.
At root level, we have 9 records. The associated splitting criterion is p g pincode = 500 046. As a result, we split the records into two subsets. Records 1, 2, 4, 8, and 9 are to the left child note and remaining to the right node. The process is repeated at every node.
R. Akerkar 11

Advantages and Shortcomings of Decision Tree Classifications

A decision tree construction process is concerned with identifying the splitting attributes and splitting criterion at every l h li i ib d li i i i level of the tree. l f h Major strengths are:

Decision tree able to generate understandable rules. They are able to handle both numerical and categorical attributes. They provide clear indication of which fields are most important for prediction or classification classification.

Weaknesses are:

The process of growing a decision tree is computationally expensive At expensive. each node, each candidate splitting field is examined before its best split can be found. Some decision tree can only deal with binary-valued target classes.

R. Akerkar

12

Iterative Dichotomizer (ID3)

Quinlan (1986) Each node corresponds to a splitting attribute Each arc is a possible value of that attribute. At each node the splitting attribute is selected to be the most informative among the attributes not yet considered in the path from the root. Entropy is used to measure how informative is a node. The algorithm uses the criterion of information gain to determine the goodness of a split. d f lit

The attribute with the greatest information gain is taken as the splitting attribute, and the data set is split for all distinct values of the attribute attribute.
R. Akerkar 13

Training Dataset
This follows an example from Quinlans ID3
The class label attribute, buys_computer, has two distinct values. Thus there are two distinct classes. (m =2) Class C1 corresponds to yes and class C2 corresponds to no no. There are 9 samples of class yes and 5 samples of class no.

age <=30 <=30 3140 >40 >40 >40 3140 <=30 <=30 >40 40 <=30 3140 3140 >40

income student credit_rating high g no fair high no excellent high no fair medium no fair low yes fair low yes excellent low yes excellent medium no fair low yes fair medium di yes f i fair medium yes excellent medium no excellent high yes fair medium no excellent

buys_computer no no yes yes yes no yes no yes yes yes yes yes no

R. Akerkar

14

Extracting Classification Rules from Trees g

Represent the knowledge in the form of IF-THEN rules One rule is created for each path from the root to a leaf Each attribute-value pair along a path forms a conjunction The leaf node holds the class p prediction Rules are easier for humans to understand
What are the rules?

R. Akerkar

15

Solution (Rules)
IF age = <=30 AND student = no THEN buys_computer = no IF age = <=30 AND student = yes THEN buys_computer = yes IF age = 3140 THEN buys_computer = yes

IF age = >40 AND credit_rating = excellent THEN buys_computer = yes IF age = <=30 AND credit_rating = fair THEN buys_computer = no

R. Akerkar

16

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 ( continuous-valued, they are g (if y 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 g p given node belong to the same class g There are no remaining attributes for further partitioning majority voting is employed for classifying the leaf There are no samples left p
R. Akerkar 17

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

Select the attribute with the highest information gain S contains si tuples of class Ci for i = {1, , m} information measures info required to classify any q y y arbitrary tuple m si si I( s1,s 2,...,s m ) log 2 s .information is encoded in bits. i 1 s entropy of attribute A with values {a1,a2,,av} f b h l {
E(A) s1 j ... smj I ( s1 j ,...,smj ) s j 1
v

information gained by branching on attribute A

Gain(A) I(s 1, s 2 ,..., sm) E(A)
R. Akerkar 18

Entropy

Entropy measures the homogeneity (purity) of a set of examples. It gives the information content of the set in terms of the class labels of the examples. Consider that you have a set of examples, S with two classes, P and N. Let the set have p instances for the class P and n instances for the class N. So the total number of instances we have is t = p + n. The view [p, n] can be seen as a class distribution of S.

The entropy for S is defined as Entropy(S) = - (p/t).log2(p/t) - (n/t).log2(n/t)

Example: Let a set of examples consists of 9 instances for class positive, and 5 instances for class negative. Answer: p = 9 and n = 5. So Entropy(S) = - (9/14).log2(9/14) - (5/14).log2(5/14) = -(0.64286)(-0.6375) - (0.35714)(-1.48557) = (0.40982) + (0.53056) = 0.940

R. Akerkar

19

Entropy y
The entropy for a completely pure set is 0 and is 1 for a set with equal occurrences f both th classes. l for b th the l i.e. Entropy[14,0] = - (14/14).log2(14/14) - (0/14).log2(0/14) = -1.log2(1) - 0 l 2(0) 1 l 2(1) 0.log2(0) = -1.0 - 0 =0 i.e. Entropy[7,7] = - (7/14).log2(7/14) - (7/14).log2(7/14) = - (0.5).log2(0.5) - (0.5).log2(0.5) = - (0.5).(-1) - (0.5).(-1) = 0.5 + 0.5 =1

R. Akerkar

20

Attribute Selection by Information Gain Computation

Class P: buys_computer = yes Class N: buys_computer = no I(p, n) = I(9, 5) =0.940 Compute the entropy for age:

5 4 E ( age ) I ( 2 ,3) I ( 4,0 ) 14 14 5 I (3, 2 ) 0 .694 14

age <=30 3040 >40 40

age <=30 <=30 3140 >40 >40 >40 3140 <=30 <=30 >40 <=30 3140 3140 >40

pi 2 4 3

ni I(pi, ni) 3 0.971 0 0 2 0.971

5 age <=30 h 5 30 has I ( 2,3) means 14 out of 14 samples, with 2 yes's

and 3 nos. Hence no s.

income student credit_rating high no fair high no excellent high no fair medium no fair low yes fair low yes excellent low yes excellent medium no fair low yes fair medium yes fair medium yes excellent medium no excellent high yes fair medium no excellent

buys_computer no no yes yes yes no yes no yes yes yes yes yes no R. Akerkar

Gain ( age ) I ( p , n ) E ( age ) 0.246

Similarly, Similarly Gain(income) 0.029
Gain( student ) 0.151 Gain(credit _ rating ) 0.048

Since, age has the highest information gain among the attributes, it is selected as the test attribute. 21

Exercise 1

The following table consists of training data from an employee database. database

Let status be the class attribute. Use the ID3 algorithm to construct a decision tree from the given data.

R. Akerkar

22

Solution 1

R. Akerkar

23

Other Attribute Selection Measures

Gini index (CART IBM IntelligentMiner) (CART,

All attributes are assumed continuous-valued Assume there exist several possible split values for each attribute May need other tools, such as clustering, to get the possible split values Can be modified for categorical attributes

R. Akerkar

24

Gini Index (IBM IntelligentMiner)

If a data set T contains examples from n classes, gini index, gini(T) is n defined as gini (T ) 1 p 2 i i
j 1 j

where pj is the relative frequency of class j in T. If a data set T is split into two subsets T1 and T2 with sizes N1 and N2 respectively, the gini index of the split data contains examples from n classes, the gini index gini(T) is defined as

gini

split

(T ) N 1 gini (T 1) N 2 gini (T 2 ) N N

The attribute provides the smallest ginisplit(T) is chosen to split the node (need to enumerate all possible splitting points for each attribute).

R. Akerkar

25

Exercise 2

R. Akerkar

26

Solution 2

SPLIT: Age <= 50 ---------------------| High | Low | Total -------------------S1 (left) | 8 | 11 | 19 S2 (right) | 11 | 10 | 21 -------------------For S1: P(high) = 8/19 = 0.42 and P(low) = 11/19 = 0.58 For S2: P(high) = 11/21 = 0.52 and P(low) = 10/21 = 0.48 Gini(S1) = 1-[0.42x0.42 + 0.58x0.58] = 1-[0.18+0.34] = 1-0.52 = 0.48 Gini(S2) = 1-[0.52x0.52 + 0.48x0.48] = 1-[0.27+0.23] = 1-0.5 = 0.5 Gini-Split(Age<=50) = 19/40 x 0.48 + 21/40 x 0.5 = 0.23 + 0.26 = 0.49 SPLIT: Salary <= 65K < ---------------------| High | Low | Total -------------------S1 (top) | 18 | 5 | 23 S2 (bottom) | 1 | 16 | 17 -------------------For S1: P(high) = 18/23 = 0.78 and P(low) = 5/23 = 0.22 For S2: P(high) = 1/17 = 0.06 and P(low) = 16/17 = 0.94 Gini(S1) = 1-[0.78x0.78 + 0.22x0.22] = 1-[0.61+0.05] = 1-0.66 = 0.34 Gini(S2) = 1-[0.06x0.06 + 0 94x0 94] = 1-[0 004+0 884] = 1-0 89 = 0 11 1-[0 06x0 06 0.94x0.94] 1-[0.004+0.884] 1-0.89 0.11 Gini-Split(Age<=50) = 23/40 x 0.34 + 17/40 x 0.11 = 0.20 + 0.05 = 0.25
R. Akerkar 27

Exercise 3

In previous exercise which is a better split of exercise, the data among the two split points? Why?

R. Akerkar

28

Solution 3

Intuitively Salary <= 65K is a better split point since it produces relatively ``pure'' partitions as opposed to Age <= 50 which pure'' 50, results in more mixed partitions (i.e., just look at the distribution of Highs and Lows in S1 and S2). More formally, let us consider the properties of the Gini index. If a partition is totally pure, i.e., has all elements from the same class, then gini(S) = 1-[1x1+0x0] = 1-1 = 0 (for two classes). On the other hand if the classes are totally mixed, i.e., both classes have equal probability then gini(S) = 1 - [0 5x0 5 + 0 5x0 5] = 1 [0 25+0 25] = 0 5 [0.5x0.5 0.5x0.5] 1-[0.25+0.25] 0.5. In other words the closer the gini value is to 0, the better the partition is. Since Salary has lower gini it is a better split. is split
R. Akerkar 29

Avoid Overfitting in Classification vo d v g C ss c o

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 y p Two approaches to avoid overfitting Prepruning: Halt tree construction earlydo 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 treeget a sequence of progressively pruned t f i l d trees Use a set of data different from the training data to decide which is the best pruned tree

R. Akerkar

30