Hypothesis on Different Data Mining Algorithms

Shraddha Deshmukh Int. Journal of Engineering Research and Applications www.ijera.com
ISSN: 2248-9622, Vol. 5, Issue 12, (Part - 3) December 2015, pp.86-91
www.ijera.com 86 | P a g e
Hypothesis on Different Data Mining Algorithms
Shraddha Deshmukh*, Swati Shinde**
*(Department of Information Technology, University of Pune, Pune - 05
** (Department of Information Technology, University of Pune, Pune - 05
ABSTRACT
In this paper, different classification algorithms for data mining are discussed. Data Mining is about
explaining the past & predicting the future by means of data analysis. Classification is a task of data mining,
which categories data based on numerical or categorical variables. To classify the data many algorithms are
proposed, out of them five algorithms are comparatively studied for data mining through classification. There are
four different classification approaches namely Frequency Table, Covariance Matrix, Similarity Functions &
Others. As work for research on classification methods, algorithms like Naive Bayesian, K Nearest Neighbors,
Decision Tree, Artificial Neural Network & Support Vector Machine are studied & examined using benchmark
datasets like Iris & Lung Cancer.
Keywords - Artificial Neural Network, Classification, Data Mining, Decision Tree, K-Nearest Neighbors, Naive
Bayesian & Support Vector Machine.
I. INTRODUCTION
Nowadays large amount of data is being
gathered and stored in databases everywhere across
the globe and it is increasing continuously. Different
organizations & research centers are having data in
terabytes. That is over 1000 Terabytes of data. So,
we need to mine those databases for better use. Data
Mining is about explaining the past & predicting the
future. Data mining is a collaborative field which
combines technologies like statistics, machine
learning, artificial intelligence & database. The
importance of data mining applications is predicted
to be huge. Many organizations have collected
tremendous data over years of operation & data
mining is process of knowledge extraction from
gathered data. The organizations are then able to use
the extracted knowledge for more clients, sales &
greater profits. This is also true in the engineering &
medical fields.
1.1 DATA MINING
Data mining is process of organising available
data in useful format. Fig.1 shows basic concept of
data mining. Basic terms in data mining are:
 Statistics: The science of collecting, classifying,
summarizing, organizing, analysing &
interpreting data.
 Artificial Intelligence: The study of computer
algorithms which simulates intelligent
behaviours for execution of special activities.
 Machine Learning: The study of computer
algorithms to grasp the experiences and use it for
computerization.
 Database: The science & technology of
collecting, storing & managing data so users can
retrieve, insert, modify or delete such data.
 Data warehousing: The science & technology of
collecting, storing & managing data with
advanced multi-dimensional reporting services in
support of the decision making processes.
 Predicting The Future: Data mining predicts the
future by means of modelling.
 Modelling: Modelling is the process in which
classification model is created to predict an
outcome.
Fig. 1. Concept of data mining
II. CLASSIFICATION
Classification is a data mining task of predicting
the value of a categorical variable (target or class) by
building a model based on one or more algebraic
and/or categorical variables (predictors or attributes).
It Classifies data based on the training set & class
labels. Examples:
 Classifying patients by their symptoms,
 Classifying goods by their properties, etc.
There are some common terms used in
classification process. Table 1 illustrates basic terms
used in classification process like pattern (records,
rows), attributes (dimensions, columns), class
(output column) and class label (tag of class):
RESEARCH ARTICLE OPEN ACCESS

TABLE -1: Terms Used in Classification
Classification is a method of data mining for
predicting the value for data instances by using
previous experiences. Since we want to predict
either a positive or a negative response, we will
build a binary classification model. Classification is
important because it helps scientists to clearly
diagnose problems, study & observe them &
organize concentrated conservation efforts. It also
assists as a way of remembering & differentiating
the types of symptoms, making predictions about
diseases of the same type, classifying the
relationship between different defects & providing
precise names for diseases.
2.1 Applications of Classification
Classification have several applications like
Medical Diagnosis, Breast Cancer Diagnosis,
Market Targeting, Image processing, Wine
Classification, Solid Classification for selection of
fertilizer, etc.
III. CLASSIFICATION ALGORITHMS
There is quite a lot of research on algorithms that
classifies data. Several approaches have been
developed for classification in data mining. Fig 2
shows hierarchy of classification algorithms:
Fig. 2. Hierarchy of classification algorithms
IV. NAÏVE BAYESIAN
4.1 Introduction to Naïve Bayesian
The Naive Bayesian (NB) method is a simple
probabilistic classifier based on Bayes Theorem
(from Bayesian statistics) with strong (naive)
independence premises which assumes that all the
features are unique. NB model is easy to build, with
no complicated iterative parameter estimation which
makes it particularly useful for very large datasets
[1].
In the NB classifiers, every feature can help
determining which topic should be appointed to a
given input value. To choose a topic for an input
value, the naive Bayes classifier begins by
evaluating the prior probability of each topic, which
is determined by checking the frequency of each
topic in the training set. The input from each feature
is then mixed with this previous probability, to
arrive at a probability estimate for each topic. If the
estimated probability is the highest is then assigned
to the testing inputs [5].
A supervised classifier is built on training
corpora containing the correct topic for each input.
The framework used by Bayesian classification is
shown in Fig.3
Fig. 3. Bayesian classification
(a) During the training, a feature extractor is used to
convert each input value to a feature set. These
feature sets, which capture the basic information
about each input that should be used to classify it,
are discussed in the next section. Pairs of feature sets
& topics are fed into the machine learning algorithm
to generate a model. (b) During the prediction, the
same feature extractor is used to convert unseen
inputs to feature sets. These feature sets are then fed
into the model, which generates predicted topics [5].
4.2 Advantages to Naive Bayesian
1. Fast to train & classify
2. Not sensitive to irrelevant features
3. Handles real & discrete data
4. Handles streaming data well
4.3 Disadvantages to Naive Bayesian
1. Assumes autonomy of aspects
2. Dependencies exist among variables (ex.
Hospital: Patients, Diseases: Diabetes, Cancer)
are not modelled by NB.
V. K NEAREST NEIGHBOR
5.1 Introduction to K-Nearest Neighbor
K nearest neighbor (KNN) is a simple method
that stores all available cases & classifies new cases
based on a similarity measure (e.g. euclidean). KNN
has been used in statistical estimation & pattern

recognition already in the beginning of 1970’s as a
non-parametric technique. The closest neighbor rule
distinguishes the classification of unknown data
point on the basis of its closest neighbor whose class
is already known [4].
The training points are selected weights
according to their distances from sample data point.
But at the same time the computational complexity
& memory requirements remain the essential things.
To overcome from memory restriction size of
dataset is minimized. For this the repeated patterns
which don’t include additional data are also
excluded from training data set. To further
strengthen the information focuses which don’t
influence the result are additionally eliminated from
training data set [4]. The NN training dataset can be
formed for utilizing different systems to boost over
memory restriction of KNN. The KNN
implementation can be done using ball tree, k-d tree,
nearest feature line (NFL), principal axis search tree
& orthogonal search tree.
Next, the tree structured training data is divided
into nodes & techniques like NFL & tunable metric
divide the training data set according to planes. The
speed of basic KNN algorithm can be increase by
using these algorithms. Consider that an object is
sampled with a set of different attributes.
5.2 Advantages to K-Nearest Neighbors
1. No assumptions about the characteristics of the
concepts to learn have to be done
2. Complex concepts can be learned by local
approximation using simple procedures
3. Very simple classifier that works well on basic
recognition problems.
5.3 Disadvantages to K-Nearest Neighbors
1. The model cannot be interpreted
2. It is computationally expensive to find the KNN
when the dataset is very large
3. It is a lazy learner; i.e. it does not learn anything
from the training data & simply uses the training
data itself for classification.
VI. DECISION TREE
6.1 Introduction to Decision Tree
A decision tree (DT) is a decision support tool
that uses a tree-like graph or model of decisions &
their possible effects, including chance event results,
assets cost & utility. It is the only way to display an
algorithm.
The decision tree is a method for information
portrayal evolved in the 60s. It can resolve the class
label of test patterns by using set value of attribute.
DT is a cycle free graph which has nodes as
attributes to support decisions. The tree branch
represents a precedence connection between the
nodes [6]. The value of a branch is an element of the
attribute value set of the branch’s parent node. The
attributes are nodes with at least two children,
because an attribute has got as many branches as the
cardinality of the value set of the actual attribute.
The root of the tree is the common ancestor attribute,
from where the classification can be started. The
leaves represent class nodes of the tree. In every
relation the class is only a child, so it is a leaf of the
tree in every case [6].
DT builds classification model in the form of a
tree. It breaks a dataset into smaller subsets and an
associated decision tree is incrementally developed.
The final result is a tree with decision nodes & leaf
nodes. A decision node has two or more branches
and leaf node represents a decision. The topmost
node also called as root node which corresponds to
the best predictor. DT can handle both categorical &
numerical data [4]. DT helps formalize the
brainstorming process so we can identify more
potential solutions.
6.2 Advantages to Decision Tree
1. Easy to interpretation.
2. Help determine worst, best & expected values for
different scenarios.
6.3 Disadvantages to Decision Tree
1. Determination can get very complex if many
values are ambiguous.
VII. ARTIFICIAL NEURAL NETWORK
6.1 Introduction to Artificial Neural Network
Artificial neural networks (ANNs) are types of
computer architecture inspired by nervous systems
of the brain & are used to approximate functions that
can depend on a large number of inputs & are
generally unknown. ANN are presented as systems
of interconnected “neurons” which can compute
values from inputs & are capable of machine
learning as well as pattern recognition due their
adaptive nature [4]. The brain basically learns from
experience. It is natural proof that some problems
that are beyond the scope of current computers are
indeed solvable by small energy efficient packages.
This brain modeling also promises a less technical
way to develop machine solutions. This new
approach to computing also provides a more
graceful degradation during system overload than its
more traditional counterparts [8].

A neural network is a massively parallel-
distributed processor made up of simple processing
units, which has a natural propensity for storing
experimental knowledge & making it available for
use. Neural network are also referred to in literature
as neuro computers, connectionist networks,
parallel-distributed processors, etc. A typical neural
network is shown in the fig. 4.Where input, hidden
& output layers are arranged in a feed forward
manner [8]. The neurons are strongly interconnected
& organized into different layers. The input layer
receives the input & the output layer produces the
final output. In general one or more hidden layers
are sandwiched in between the two [4]. This
structure makes it impossible to forecast or know the
exact flow of data.
Input Layer Hidden Layer Output Layer
Fig. 4. A simple neural network
ANN typically starts out with randomized
weights for all their neurons. This means that
initially they must be trained to solve the particular
problem for which they are proposed. During the
training period, we can evaluate whether the ANN’s
output is correct by observing pattern. If it’s correct
the neural weightings that produced that output are
reinforced; if the output is incorrect, those
weightings responsible can be diminished [4]. An
ANN is useful in a variety of real-world applications
such as visual pattern recognition, speech
recognition and programs for text-to-speech that deal
with complex often incomplete data.
6.2 Advantages to Artificial Neural Network
1. It is easy to use, with few parameters to adjust.
2. A neural network learns & reprogramming is not
needed.
3. Applicable to a wide range of problems in real
life.
6.3 Disadvantages to Artificial Neural
Network
1. Requires high processing time if neural network
is large.
2. Learning can be slow.
VIII. SUPPORT VECTOR MACHINE
8.1 Introduction to Support Vector Machine
A Support Vector Machine (SVM) performs
classification by finding the hyperplane that
maximizes the margin between the two classes. The
vectors (cases) that define the hyperplane are the
support vectors.
The beauty of SVM is that if the data is linearly
separable, there is a unique global minimum value.
An ideal SVM analysis should produce a hyperplane
that completely separates the vectors (cases) into
two non-overlapping classes. However, perfect
separation may not be possible, or it may result in a
model with so many cases that the model does not
classify correctly. In this situation SVM finds the
hyperplane that maximizes the margin & minimizes
the misclassifications [4].The algorithm tries to
maintain the slack variable to zero while maximizing
margin. However, it does not minimize the number
of misclassifications (NP-complete problem) but the
sum of distances from the margin hyperplanes [9].
The simplest way to separate two groups of data
is with a straight line (1 dimension), flat plane (2
dimensions) or an N-dimensional hyperplane as
shown in Fig. 5.
Fig. 5. Hyperplane in SVM
The main reason you would want to use an SVM
instead of a Logistic Regression is because problem
might not be linearly separable and if you are in a
highly dimensional space [9].
8.2 Advantages to Support Vector Machine
1. Most robust & accurate classification technique
for binary problems.
2. Memory-intensive.
8.3 Disadvantages to Support Vector
Machine
1. It can be painfully inefficient to train
2. High complexity & extensive memory
requirements for classification in many cases.
3. Supports only binary classification.

IX. COMPARATIVE STUDY OF ALGORITHMS
As per comparative analysis from Table 2 we can
say that, ANN is better for data classification. Neural
network allows learning from experiences &
supports decision making, classification, Pattern
recognition, etc. Neural networks often exhibit
patterns similar to those exhibited by humans.
However this is more of interest in cognitive
sciences than for practical examples.
TABLE -2: Comparative Study of Algorithms
Algo. Approach Features Flaws
NB Frequency
Table
 Simple to
implement.
 Great
Computatio
nal
efficiency &
classificatio
n rate.
 It predicts
accurate
results for
most of the
classificatio
n &
prediction
problems.
 The
precision of
algorithm
decreases if
the amount
of data is
less.
 For
obtaining
good results
it requires a
very large
number of
records.
KNN Similarity
Function
 Classes need
not be
linearly
separable.
 Zero cost of
the learning
process.
 Well suited
for
multimodal
classes.
 Time to find
the nearest
Neighbors
in a large
training data
set can be
excessive.
 Performance
of algorithm
depends on
the number
of
dimensions
used
DT Frequency
Table
 It produces
the more
accuracy
result than
the C4.5
algorithm.
 Detection
rate is
increase &
space
consumption
is reduced.
 Requires
large
searching
time.
 Sometimes
it may
generate very
long rules
which are
very hard to
prune.
 Requires
large amount
of memory
to store tree.
ANN Others  It is easy to
use &
implement,
with few
parameters
to adjust.
 A neural
network
learns &
reprogrammi
ng is not
needed.
 Applicable
to a wide
range of
problems in
real life.
 Requires
high
processing
time if
neural
network is
large.
 Learning
can be
slow.
SVM Others  High
accuracy.
 Work well
even if data
is not
linearly
separable in
the base
feature
space.
 Speed &
size
requiremen
t both in
training &
testing is
more.
 High
complexit
y &
extensive
memory
requireme
nts for
classificati
on in
many
cases.
X. EXPERIMENTAL ANALYSIS
To examine all studied methods Lung Cancer &
Iris benchmark datasets are used. Results are for
100% training & 100% testing scenario. Result
analysis on the basis of accuracy is given in Table 3.
Accuracy is calculated as:
Total number of correctly classified data
Accuracy = -------------------------------------------------
Total number of data
TABLE -3: Result Analysis of Methods Using
WEKA Tool
Iris Lung Cancer
NB 94.7% 97.4%
KNN 94.0% 97.0%
DT 95.0% 98.0%
SVM 65.0% 98.5%
ANN 98.3% 100%

0
20
40
60
80
100
NB KNN DT SVM ANN
Iris
Lung Cancer
Fig 6: Chart of Result Analysis
XI. CONCLUSION
In this paper, we have studied five different
classification methods based on approaches like,
frequency tables, covariance matrix, similarity
functions & others. Those algorithms are NB, KNN,
DT, ANN & SVM. As per comparative study in
between all algorithms we reached to conclusion that
ANN is most suitable & efficient technique for
Classification. ANNs are considered as simplified
mathematical models of human brain & they
function as parallel distributed computing networks.
ANNs are universal function approximates, &
usually they deliver good performance in
applications. ANN has generalization ability as well
as learnability. It is easy to use, implement &
applicable to real world problems.
There is a huge scope in this area of
classification by using different methods of ANN
like Fuzzy with ANN, Neuro-fuzzy, Genetic
Approach, etc. in Artificial Neural Network.
REFERENCES
Journal Papers:
[1] M Ozaki, Y. Adachi, Y. Iwahori, and N. Ishii, Application
of fuzzy theory to writer recognition of Chinese characters,
International Journal of Modeling and Simulation, 18(2),
1998, 112-116.
[2] R. Andrews, J. Diederich & A. B., Tickle, “Survey &
critique of techniques for extracting rules from trained
artificial neural networks,” Knowledge Based System, vol. 8,
no. 6, pp. 373-389, 1995.
[3] Rashedur M. Rahman, Farhana Afroz, “Comparison of
Various Classification Techniques Using Different Data
Mining Tools for Diabetes Diagnosis”, Journal of Software
Engineering & Applications, 6, 85-97, 2013.
[4] Wang Xin, Yu Hongliang, Zhang Lin, Huang Chaoming,
Duan Jing, “Improved Naive Bayesian Classifier Method &
the Application in Diesel Engine Valve Fault Diagnostic”,
Third International Conference On Measuring Technology
& Mechatronics Automation, 2011.
[5] Sagar S. Nikam, “A Comparative Study Of Classification
Techniques In Data Mining Algorithms”, Oriental Journal
Of Computer Science & Technology, April 2015.
[6] Zolboo Damiran, Khuder Altangerelt, “Text Classification
Experiments On Mongolian Language”, IEEE Conference,
Jul 2013.
[7] Zolboo Damiran, Khuder Altangerel, “Author Identification:
An Experiment based on Mongolian Literature using
Decision Tree”, IEEE Conference, 2013.
[8] Essaid el Haji, Abdellah Azmani, Mohm el Harzli, “A
pairing individual-trades system , using KNN method”,
IEEE Conference, 2014.
[9] Kumar Abhishek, Abhay Kumar, Rajeev Ranjan, Sarthak
K., “A Rainfall Prediction Model using Artificial Neural
Network”, IEEE Conference, 2012.
[10] Erlin, Unang Rio, Rahmiati, “Text Message Categorization
of Collaborative Learning Skills in Online Discussion Using
SVM”, IEEE Conference, 2013.

Hypothesis on Different Data Mining Algorithms

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Hypothesis on Different Data Mining Algorithms (20)

Recently uploaded (20)

Hypothesis on Different Data Mining Algorithms