IRJET- The Machine Learning: The method of Artificial Intelligence

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 04 | Apr-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 235
The Machine Learning: The method of Artificial Intelligence
Ashis Kumar Ratha1,Nisha Agrawal2, Amisha Ananya Sikandar3
1Asst.Prof, Department of Computer Science & Engg, VIT, Bargarh, Odisha ,INDIA
2,3Student Researcher, Department of Computer Science & Engg, VIT, Bargarh Odisha ,INDIA
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract: In classification problem learning and decision
making is at the core level of argument as well as artificial
aspects. So scientists introduce machine learning which is
widely used in artificial intelligence. Artificial intelligence
planning systems have become an important tool for
automating a wide variety of tasks. Machine Learning
techniques enable a planning system to automatically
acquired search control knowledge for different applications.
In the field of robotics machine learning plays an important
role, it helps in taking decision and increase the efficiency of
the machine. Machine learning is used in much application
which is the principle concept for intelligence system which
assist to the ingenious introduction and advanced concepts of
artificial intelligence.
1. INTRODUCTION
A type of artificial intelligence, which allows software
applications whose accuracy in predicting output without
explicitly programmed, is machine learning.Dataminingand
predictive modeling carries same process as machine
learning. In this study of biological and artificial vision,
learning used as a key.
To comprehend the virtual environment relating to the
machine understanding different algorithm are introduced
for avoiding to build the heavy machine having explicit
programming .For taking independent decision different
algorithm are implemented in different machines. A huge
number of data sets being given to classify and based on
these data sets it do some processing and triesto predict the
result .pattern recognition is the innate process of matching
information from the environment withinformationstorein
memory. Pattern recognition is closely related to top down
perception. In both casesknowledge and expectationareuse
in interpret information.
Pattern recognition involves detection of repeat
characteristic, occurrences or some other attribute and this
basic way to make sense of the word. In other way the
constant attempt for identifying environmental information
that matches the internal information.
In other words a branch of machine learning which used by
many algorithms for getting optimized decisions is pattern
recognition.
Fig1 : The Machine Learning Mechanism
RELATED WORKS:
Sally Goldman et.al[1] gave the practical learning scenarios
in which small amount of labeled data with a huge unlabeled
data presented a co training strategy for using unlabeled
data for improving the standard supervised learning
algorithm. According to her assumption there are two types
of hypothesis which defines the partitioning of instance
space. Eg: The instance space with one equivalent class
defined per tree in decision tree partition. The conclusion
she gave was two supervised learning algorithm canbeused
successfully label data for each other.
Zoubin Ghahramani et.al[2] Theoverviewofunsupervised
learning from statistical modeling is provided by him .He
concludes that unsupervised learning canbemotivatedfrom
information theoretic and Bayesian principles. He further
wind up that statistics provide a coherent framework for
learning from data and reasoning and he divulge thetypesof
model likegraphicalmodel which play a vitalroleinlearning
for handling of different kinds of data.
Rich Caruana et.al[3] The comparison between ten
supervised learning methods has studied by him in
supervised learning method introducedinlastdecade.These
methods includes SVMS ,neural nets, logistic regression,
naïvebayes, memory based learning,randomforest,decision
trees, bagged trees ,boosted tree and boosted stumps. To
evaluate the learning methodsthey studied and examinethe
effect that calibrating the models through plat scaling and
isotonic regression.

Niklas lavesson et.al[4] According to him performance
is often only measured in terms of accuracy. Through cross
validation tests, however some researches have given a
different approach for evaluation of supervised learning. i.e
measure function a limitation of current measure functions
is that they can only handle two dimensional instance
spaces. They present design and implementation of a
generalized multidimensional measure function and
demonstrate its use through a set of experiments. The result
indicate that there are cases for which measure functions
may be able to capture aspects of performance that cannot
be captured by cross validation tests .Final result will be,
they investigate the impact of learning algorithmparameter
tuning.
Yogowati Praharsi et.al[5] The three supervised
learning methods as k-nearest neighbor(k-NN),support
vector data description(SVDD) and support vector
machine(SVM) are approached by him because they do not
suffer from the complexity of introducing a new class ,and
further used for data description and classification. The
output show that feature selection based on mean
information gain and a standard deviation threshold can be
considered as a substitute for forward selection.
PROBLEM FACED IN LEARNING:
As so many decisions are made, learning considered as a
complex process depending upon machine to machine as
well as algorithm to algorithm. From understanding a
problem to responding, so many issues make a complex
situation to respond for a machine, so it affects the learning
process.
Perception defines how the machine perceives, so machine
should also aim different types of challenges and
environment to face. Though different inputsresultdifferent
output, machine should considered only the optimized and
appropriate output.
Problems faced during learning process are as
follows:
Bias:Any error occur in learning algorithmistermedasbias.
The problem faced during simultaneously minimizing two
sources of error which prevent algorithm of supervised
learning.
Noise: The unwanted data and the imperfection of data are
now common in real world situation. The noise exist in the
data are degrade the learning process but one of the
properties of learning algorithm is to handling noisy data in
all form.
Pattern Recognition: The next problem termed as pattern
recognition, which aim is to providing reasonable answer of
all inputs and performsthe matching operation forallinputs
and performs the matching operationforallinputsaccording
to their statistical variation.
As machine is well known of mathematical models (square,
rectangle, circle, etc), but it is also true that It become
different for machine to process those inputs having
different values.
Both the inputs and outputs are perceived in supervised
learning .For responding all the inputs the algorithm has to
generate all the inputs the algorithm has to generate all the
training data from supervised learning.
When any agent is given immediate feedback supervised
learning of action occurs. For solving any given problem
using supervised learning, some steps to be carried out are:
1. Determination of training example and its type
2. Collecting training set
3. Knowledge of input feature of learned function
4. Determination of structure of learning function
5. Completion of decision, to run learningalgorithmbasedon
gathered set of data
6.Optimising the accuracy of learned function and
performance of learning function and also performance
should again measured on the set which are different from
training set .
Fig 2: Supervised Learning Algorithm
The two category of supervised learning are:
1. Grouping of responses having only truth values(true or
false).
2. Retroversion of responses having real values.

In supervised learning inputs are received but failed to
obtain supervised target outputs and rewards from its
environment .Though it failed but it is possible for
developing a formal framework for unsupervised learning
like clustering and dimensionality reduction.
Fig 3: Working Mechanism of Supervised Algorithm
Algorithm for unsupervised learning:
1 .HIERARCHIAL CLUSTERING:
Hierarchial clustering is a method of cluster analysis in
which we look forward to build a hierarchy of cluster. The
aim of this algorithm is to make a multilevel hierarchy of
cluster by creating cluster tree.
Inputs: objects represented as vectors
Output: a hierarchy of associations represented as a “
Dendogram”
Algorithm:
1. hclust (D,:set of instances ):tree
2. var:C:/*set of clusters*/
3. M/*matrix containing distance between 2
clusters*/
4. For each dԐD}do
5. Make a as leaf node in C
6. Done
7. For each pair a,bԐC do
8. Ma,b←d(a,b)
9. Done
10.While( not all instances in one cluster)do
11.Find the most similar pair of cluster in M
12.Merge these two cluster into one cluster
13.Update M to reflect the merge operation
14.Done
15.Return C
2. K-MEANS CLUSTERING
In data mining, a method of vector quantization for cluster
analysis is used i.e known as k-means clustering. The aim of
k-means clustering is that the partitioning of n observation
belongs to the nearest cluster as prototype.
ALGORITHM:
1.K-means ((X={d1…..dn} Rm,k):2R)
2. C:2R /* μ a set of clusters*/
3.d:RxRm->R/*distance function*/
4.μ:2R->R/*μ computers the mean of a cluster*/
5. select C with k initial centers f1,….fk
6. while stopping criterion not true do
7.for all clusters cj Ԑ C do
8.cj←{diᵾf1d(di,fj) ≤d(di,f1)}
9. done
10.for all means fj do
11. fj←μ(cj)
12. done
Fig 4: Working Mechanism of Unsupervised Learning
Model
CONCLUSION
The investigation of performance measurement of learning
algorithm has been studied. This is the complicated query
with many aspects. Some issues like analyzing evaluation
methodsand the metrics which measure performance anda
frame work which describe the methods in a structural
way.The conclusion that we made by the analysis is that the
measurement of classifier performance is calculate by
accuracy like in cross validation test .some general methods
are used to evaluate any classifier or any algorithm by the
structure of representation, while other methods have
restricted to any certain algorithm of representation. The
visualization of classifier performance is requiredbecauseof
the method doesn’t work like a function returning a
performance as result. Measure based evaluation for
measuring classifier performance hasalso beeninvestigated
and we provide factual experiments results that strengthen
earlier publication of theoretical arguments for measure
based evaluation. This experiment was capable of
differentiate between classifiersthat we were acquantedvia
accuracy but different in complexity.

REFERENCES
[1] Sally Goldman; Yan Zhou, "Enhancing Supervised
Learning with Unlabeled Data", Department of Computer
Science, Washington University, St.Louis, MO 63130 USA.
[2] Zoubin Ghahramani,"Unsupervised Learning",Gatsby
Computational Neuroscience Unit, University College
London, UK.
[3] Rich Caruana; Alexandru Niculescu- Mizil,"An Empirical
Comparison of Supervised Learning Algorithms",
Department of Computer Science, Cornell University, Ithaca,
NY 14853 USA
[4] Niklas Lavesson,"Evaluation and Analysis of Supervised
Learning Algorithms and Classifiers", Blekinge Institute of
Technology Licentiate Dissertation Series No 2006:04,ISSN
1650-2140,ISBN 91-7295-083-8
[5] Yugowati Praharsi; Shaou-Gang Miaou; Hui-Ming Wee,”
Supervised learning approaches and feature selection - a
case study in diabetes”, International Journal of Data
Analysis Techniques and Strategies 2013 - Vol. 5,No.3 pp.
323 – 337
[6] Andrew Ng,"Deep Learning And Unsupervised","Genetic
Learning Algorithms","ReinforcementLearningandControl",
Department of Computer Science,Stanford University,450
Serra Mall, CA 94305, USA.
[7] Bing Liu, "Supervised Learning", Department of
Computer Science, University of IllinoisatChicago(UIC),851
S. Morgan Street, Chicago
[8]Niklas Lavesson,"Evaluation and Analysis of Supervised
Learning Algorithms and Classifiers", Blekinge Institute of
Technology Licentiate Dissertation Series No 2006:04,ISSN
1650-2140,ISBN 91-7295-083-8.
[9]Rich Caruana; Alexandru Niculescu- Mizil,"An Empirical
Comparison of Supervised Learning Algorithms",
Department of Computer Science, Cornell University, Ithaca,
NY 14853 USA
[10] Peter Norvig; Stuart Russell,"Artificial Intelligence: A
Modern Approach".
BIOGRAPHIES
Mr.A.K.Ratha is working as an
assistant professor in department
of CSE, Vikash Institute of
Technology, Bargarh. He has done
MCA in 2007 from Sambalpur
University and M.Tech. in CSE with
Information Securityspecialization
in 2012 from KIIT University,
Bhubaneswar. He has 10 years of
experience in teaching
engineering. His research area is
Natural Language Processing,
Application Programming &
Machine Learning.
Ms. Nisha Agrawal is perusing
B.tech(CSE) from Vikash Institute of
Technology, Bargarh and presently is
in Fourth Year. Her research interests
are Database management systemand
Operating System.
Ms. Amisha Ananya Sikandar is
perusing B.tech(CSE) from Vikash
Institute of Technology, Bargarh and
presently is in Fourth Year. Her
research interests are Operating
System and Data Structure

IRJET- The Machine Learning: The method of Artificial Intelligence

More Related Content

What's hot (20)

Similar to IRJET- The Machine Learning: The method of Artificial Intelligence (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- The Machine Learning: The method of Artificial Intelligence