PRESENTATION ON

NAÏVE BAYESIAN CLASSIFICATION

Presented By:

http://ashrafsau.blogspot.in/
Ashraf Uddin
Sujit Singh
Chetanya Pratap Singh
South Asian University
(Master of Computer Application)
http://ashrafsau.blogspot.in/
 OUTLINE

 Introduction to Bayesian Classification

 Bayes Theorem
 Naïve Bayes Classifier

http://ashrafsau.blogspot.in/
 Classification Example

 Text Classification – an Application

 Comparison with other classifiers

 Advantages and disadvantages
 Conclusions
 CLASSIFICATION
 Classification:
 predicts categorical class labels

http://ashrafsau.blogspot.in/
 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

 Typical Applications
 credit approval
 target marketing
 medical diagnosis
 treatment effectiveness analysis
 A TWO STEP PROCESS
 Model construction: describing a set of predetermined
classes
 Each tuple/sample is assumed to belong to a predefined

http://ashrafsau.blogspot.in/
class, as determined by the class label attribute
 The set of tuples used for model construction: 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
 INTRODUCTION TO BAYESIAN CLASSIFICATION

 What is it ?

http://ashrafsau.blogspot.in/
 Statistical method for classification.
 Supervised Learning Method.
 Assumes an underlying probabilistic model, the Bayes
theorem.
 Can solve problems involving both categorical and
continuous valued attributes.
 Named after Thomas Bayes, who proposed the Bayes
Theorem.
http://ashrafsau.blogspot.in/
 THE BAYES THEOREM

 The Bayes Theorem:

 P(H|X)= P(X|H) P(H)/ P(X)

http://ashrafsau.blogspot.in/
 P(H|X) : Probability that the customer will buy a computer given that
we know his age, credit rating and income. (Posterior Probability of
H)
 P(H) : Probability that the customer will buy a computer regardless
of age, credit rating, income (Prior Probability of H)
 P(X|H) : Probability that the customer is 35 yrs old, have fair credit
rating and earns $40,000, given that he has bought our computer
(Posterior Probability of X)
 P(X) : Probability that a person from our set of customers is 35 yrs
old, have fair credit rating and earns $40,000. (Prior Probability of X)
 http://ashrafsau.blogspot.in/
BAYESIAN CLASSIFIER
 http://ashrafsau.blogspot.in/
NAÏVE BAYESIAN CLASSIFIER...
 http://ashrafsau.blogspot.in/
NAÏVE BAYESIAN CLASSIFIER...
 “ZERO” PROBLEM
 What if there is a class, Ci, and X has an attribute
value, xk, such that none of the samples in Ci has

http://ashrafsau.blogspot.in/
that attribute value?

 In that case P(xk|Ci) = 0, which results in P(X|Ci) =

0 even though P(xk|Ci) for all the other attributes in
X may be large.
 NUMERICAL UNDERFLOW
 When p(x|Y ) is often a very small number: the
probability of observing any particular high-

http://ashrafsau.blogspot.in/
dimensional vector is small.
 This can lead to numerical under flow.
 http://ashrafsau.blogspot.in/
LOG SUM-EXP TRICK
 BASIC ASSUMPTION
 The Naïve Bayes assumption is that all the features
are conditionally independent given the class label.

http://ashrafsau.blogspot.in/
 Even though this is usually false (since features are
usually dependent)
EXAMPLE: CLASS-LABELED TRAINING TUPLES
FROM THE CUSTOMER DATABASE

http://ashrafsau.blogspot.in/
 http://ashrafsau.blogspot.in/
EXAMPLE…
 http://ashrafsau.blogspot.in/
EXAMPLE…
 USES OF NAÏVE BAYES CLASSIFICATION

 Text Classification
 Spam Filtering

http://ashrafsau.blogspot.in/
 Hybrid Recommender System
 Recommender Systems apply machine learning and
data mining techniques for filtering unseen
information and can predict whether a user would like
a given resource
 Online Application
 Simple Emotion Modeling
TEXT CLASSIFICATION – AN
APPLICATION OF NAIVE BAYES
CLASSIFIER

http://ashrafsau.blogspot.in/
WHY TEXT CLASSIFICATION?

 Learning which articles are of interest

http://ashrafsau.blogspot.in/
 Classify web pages by topic

 Information extraction

 Internet filters
EXAMPLES OF TEXT CLASSIFICATION

 CLASSES=BINARY
 “spam” / “not spam”

http://ashrafsau.blogspot.in/
 CLASSES =TOPICS
 “finance” / “sports” / “politics”

 CLASSES =OPINION
 “like” / “hate” / “neutral”

 CLASSES =TOPICS
 “AI” / “Theory” / “Graphics”

 CLASSES =AUTHOR
 “Shakespeare” / “Marlowe” / “Ben Jonson”
EXAMPLES OF TEXT CLASSIFICATION
 Classify news stories as
world, business, SciTech, Sports ,Health etc
 Classify email as spam / not spam

http://ashrafsau.blogspot.in/
 Classify business names by industry
 Classify email to tech stuff as Mac, windows etc
 Classify pdf files as research , other
 Classify movie reviews as
favorable, unfavorable, neutral
 Classify documents
 Classify technical papers as
Interesting, Uninteresting
 Classify Jokes as Funny, Not Funny
 Classify web sites of companies by Standard
Industrial Classification (SIC)
NAÏVE BAYES APPROACH
 Build the Vocabulary as the list of all distinct words that
appear in all the documents of the training set.

http://ashrafsau.blogspot.in/
 Remove stop words and markings
 The words in the vocabulary become the
attributes, assuming that classification is independent of
the positions of the words
 Each document in the training set becomes a record
with frequencies for each word in the Vocabulary.
 Train the classifier based on the training data set, by
computing the prior probabilities for each class and
attributes.
 Evaluate the results on Test data
REPRESENTING TEXT: A LIST OF WORDS

http://ashrafsau.blogspot.in/
 Common Refinements: Remove Stop
Words, Symbols
TEXT CLASSIFICATION ALGORITHM:
NAÏVE BAYES

http://ashrafsau.blogspot.in/
 Tct – Number of particular word in particular class
 Tct’ – Number of total words in particular class

 B’ – Number of distinct words in all class

 http://ashrafsau.blogspot.in/
EXAMPLE
EXAMPLE CONT…

http://ashrafsau.blogspot.in/
oProbability of Yes Class is more than that of No
Class, Hence this document will go into Yes Class.
Advantages and Disadvantages of Naïve Bayes
Advantages :
Easy to implement
Requires a small amount of training data to estimate
the parameters

http://ashrafsau.blogspot.in/
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 modelled by Naïve
Bayesian Classifier
An extension of Naive Bayes for delivering robust
classifications

•Naive assumption (statistical independent. of the features

given the class)

http://ashrafsau.blogspot.in/
•NBC computes a single posterior distribution.
•However, the most probable class might depend on the
chosen prior, especially on small data sets.
•Prior-dependent classifications might be weak.
Solution via set of probabilities:
•Robust Bayes Classifier (Ramoni and Sebastiani, 2001)
•Naive Credal Classifier (Zaffalon, 2001)
•Violation of Independence Assumption

http://ashrafsau.blogspot.in/
•Zero conditional probability Problem
VIOLATION OF INDEPENDENCE ASSUMPTION

 Naive Bayesian classifiers assume that the effect of

an attribute value on a given class is independent

http://ashrafsau.blogspot.in/
of the values of the other attributes. This
assumption is called class conditional
independence. It is made to simplify the
computations involved and, in this sense, is
considered “naive.”
 IMPROVEMENT

 Bayesian belief network are graphical

models, which unlike naïve Bayesian

http://ashrafsau.blogspot.in/
classifiers, allow the representation of
dependencies among subsets of attributes.

 Bayesian belief networks can also be used for

classification.
 ZERO CONDITIONAL PROBABILITY PROBLEM
 If a given class and feature value never occur
together in the training set then the frequency-

http://ashrafsau.blogspot.in/
based probability estimate will be zero.

 This is problematic since it will wipe out all

information in the other probabilities when they are
multiplied.

 It is therefore often desirable to incorporate a small-

sample correction in all probability estimates such
that no probability is ever set to be exactly zero.
 CORRECTION

 To eliminate zeros, we use add-one or Laplace

smoothing, which simply adds

http://ashrafsau.blogspot.in/
one to each count
 EXAMPLE
 Suppose that for the class buys computer D (yes) in some training
database, D, containing 1000 tuples.
 we have 0 tuples with income D low,
 990 tuples with income D medium, and
 10 tuples with income D high.

http://ashrafsau.blogspot.in/
 The probabilities of these events, without the Laplacian correction, are
0, 0.990 (from 990/1000), and 0.010 (from 10/1000), respectively.
 Using the Laplacian correction for the three quantities, we pretend that
we have 1 more tuple for each income-value pair. In this way, we instead
obtain the following probabilities :

respectively. The “corrected” probability estimates are close to their

“uncorrected” counterparts, yet the zero probability value is avoided.
 Remarks on the Naive Bayesian Classifier
•Studies comparing classification algorithms have found that
the naive Bayesian classifier to be comparable in

http://ashrafsau.blogspot.in/
performance with decision tree and selected neural network
classifiers.

•Bayesian classifiers have also exhibited high accuracy and

speed when applied to large databases.
 Conclusions
The naive Bayes model is tremendously appealing because of its
simplicity, elegance, and robustness.
 It is one of the oldest formal classification algorithms, and yet even in

http://ashrafsau.blogspot.in/
its simplest form it is often surprisingly effective.
It is widely used in areas such as text classification and spam filtering.
A large number of modifications have been introduced, by the
statistical, data mining, machine learning, and pattern recognition
communities, in an attempt to make it more flexible.
 but some one has to recognize that such modifications are
necessarily complications, which detract from its basic simplicity.
 REFERENCES

 http://en.wikipedia.org/wiki/Naive_Bayes_classifier
 http://www.cs.cmu.edu/afs/cs.cmu.edu/project/theo-

http://ashrafsau.blogspot.in/
20/www/mlbook/ch6.pdf
 Data Mining: Concepts and Techniques, 3rd
Edition, Han & Kamber & Pei ISBN: 9780123
814791