International Journal of Innovative Technology and Exploring Engineering (IJITEE)

ISSN: 2278-3075, Volume-8, Issue-9S, July 2019

An Insight into Machine Learning Techniques for

Predictive Analysis and Feature Selection
Deepti Aggarwal, Vikram Bali, Sonu Mittal

Abstract - Predictive analysis comprises a vast variety of

statistical techniques like “machine learning”, “predictive
modelling” and “data mining” and uses current and historical
statistics to predict future outcomes. It is used in both business Data
and educational domain with equal applicability.This paper aims Collection
to give an overview of the top work done so far in this field. We
have briefed on classical as well as latest approaches Model Data Pre-
(using“machine learning”) in predictive analysis. Main aspects Evaluation processing
like feature selection and algorithm selection along with
corresponding application is explained. Some of the most quoted
papers in this field along with their objectives are listed in a Predictive
Analysis Life
table. This paper can give a good heads up to whoever wants to
Cycle
know and use predictive analysis for his academic or business
application.
Model Feature
Deployment Selection
Index Terms:Classification, Clustering, Feature Selection,
Machine Learning, Predictive Analysis,Regression
Predictive
Modelling
I. INTRODUCTION

Predictive analytics is driven by predictive modelling.

Predictive analytics use machine learning algorithms [2][3]
Fig. 1: Predictive analysis life cycle
and hence both go hand-in-hand. Predictive models can be
trained on a data set to respond to a new set of data
II. RELATED WORK
or values. These outcomes may be the possible changes in
the market or may be behaviour of a customer. It helps to
Cai et al.discussed various methods and evaluation
estimate the future occurrences on the basis of past
parameters for feature selection, that are widely applied in
occurrences. Predictive analytics is frequently used for
various classification and clustering machine learning
security, marketing, risk, operations and fraud detection problems[1].Sheikhpoursurveyed “semi-supervised feature
[15][16][20].The two types of predictive models are: 1) selection methods” and presented two taxonomies of these
Classification models, used for predicting class membership based on two different perspectives that represented the
2) Regression models, used for predicting a number. The hierarchical structure of “semi-supervised feature selection
predictive models consist of algorithms, which are used to methods”[6]. Chandrashekar and Sahin demonstrated the
perform data mining and statistical analysis to determine the applicability of techniques used for feature selection and
patterns and trends in data. provided a brief introduction to variable elimination which
The life cycle for predictive analysis is shown in fig. 1. can be applied to a wide array of machine learning
The first step is to collect the data on which modelling has problems and also talked about Wrapper, Filter and
to be done. The data pre-processing includes data cleaning Embedded methods[17].
and data integration. The influential features are then
selected out the total features using feature selection Miao and Niu applied feature selection algorithms on
algorithms. The model is constructed using one or more machine learning problems and found that “unsupervised
machine learning algorithms and deployed. The resultant feature selection algorithms” improves the performance of
models are then evaluated based on multiple performance clustering algorithms[10]. Liu et al.proposed “a new
metrics. statistical measure named as LW-index which could replace
the expensive cross-validation scheme to evaluate the
Revised Manuscript Received on July 02, 2019. feature subset. Then, a new feature selection method, which
Deepti Aggarwal, Research Scholar, School of Computer and System
Sciences, Jaipur National University, Jaipur, India,
is the combination of the proposed LW-index with
aggarwal.deepti@gmail.com Sequence Forward Search algorithm (SFS-LW), is
Dr. Vikram Bali, Professor & Head, Department of Computer Science presented in this paper”[8].
and Engineering, JSS Academy of Technical Education, Noida, India,
vikramgcet@gmail.com
Dr. Sonu Mittal, Associate Professor, School of Computer and System Qiu et al.presented a literature survey of the latest
Sciences, Jaipur National University, Jaipur, India, advancements and learning methods in the field of
dr.sonumittal@jnujaipur.ac.in “machine learning” for “big
data processing”. The
authors investigated the

Published By:
Retrieval Number: I10550789S19/19©BEIESP 342 Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.I1055.0789S19 & Sciences Publication
 An Insight Into Machine Learning Techniques For Predictive Analysis And Feature Selection

connections between machine learning and “signal learning techniques and statistical techniques and concluded
processing techniques for big data processing”[11]. Landset that the machine learning techniques can predict software
et al. discussed the merits and demerits of three different fault proneness[14].
processing paradigms and also compared the engines that
implement them[12]. The work related to predictive modelling, along with their
objective and techniques involved are summarized in Table
Lykourentzou et al. proposed a dropout prediction method I.
for e-learning courses using “feed-forward neural
networks”, “support vector machines” and “probabilistic
ensemble simplified fuzzy ARTMAP”[19]. Ruchika
Malhotra compared the performance of different machine

Table I: Summarization of papers based on different machine learning techniques

S. No. Technique Paper Title Year Author Objective

1 Clustering “A new soft 2019 “Hoang Nguyen, Xuan- To develop a hybrid model based on
computing Nam Bui, Quang-Hieu “Hierarchical K-means clustering (HKM)”
model for Tran, Ngoc-Luan Mai” and “Cubist algorithm (CA)”, code name
estimating and “HKM-CA model” to predict “blast-induced
controlling PPV on-site to control undesirable effects on
blast-produced the surrounding environment”. This model
ground vibration improved the accuracy of the CA model with
based on a RMSE of 0.475 and MAE of 0.373.
Hierarchical K-
means clustering
and Cubist
algorithms”
2 Linear “A machine 2018 “Ahmed Abdelaziz, To propose a model for Health Care Services
Regression learning model Mohamed Elhoseny, based on cloud environment using “Parallel
and Neural for improving Ahmed S. Salama, A.M. Particle Swarm Optimization (PPSO)” in
Network healthcare Riad” order to optimize the Virtual Machine’s
services on selection and for chronic kidney disease
cloud computing diagnosis
environment”

3 Logistic “Data mining 2018 “Concepción Burgos, To predict if a student will drop out in his
Regression for modelling María L. Campanario, course or not
students’ David de la Peña, Juan
performance: A A. Lara, David Lizcano,
tutoring action María A. Martínez”
plan to prevent
academic
dropout”

4 Logistic “A new hybrid 2018 “Arno De Caigny, To design a new algorithm called “logit leaf
Regression classification Kristof Coussement, model” which is a hybrid of “Decision trees”
algorithm for Koen W. De Bock” and “logistic regression” algorithms in
customer churn customer churn prediction
prediction based
on logistic
regression and
decision trees”

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: I10550789S19/19©BEIESP 343 & Sciences Publication
DOI: 10.35940/ijitee.I1055.0789S19
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8, Issue-9S, July 2019
5 Clustering “Cloud based 2018 “Shakeel, P.M., Baskar, To compare the efficiency of different
framework for S., Dhulipala” clustering techniques in order to diagnose
diagnosis of which gender and age group are mostly
diabetes mellitus affected by diabetes.
using K-means
clustering”

6 Decision “A new hybrid 2018 “Arno De Caigny, To propose a hybrid algorithm, “the logit
Tree classification Kristof Coussement, leaf model (LLM)”, based on “Decision
algorithm for Koen W. De Bock” trees” and “Logistic regression” for customer
customer churn churn prediction.
prediction based
on logistic
regression and
decision trees”

7 Logistic “Early detection 2017 “Anne-Sophie Hoffait, To design the prediction algorithms with
Regression of university Michaël Schyns” high accuracy using “random forest”,
students with “logistic regression” and “artificial neural
potential network” algorithms.
difficulties”

8 Clustering “Trainable 2017 “Ignacio Arganda- To introduce the “Trainable Weka

Weka Carreras, Verena Segmentation (TWS), a machine learning
Segmentation: a Kaynig, Curtis Rueden, tool that leverages a limited number of
machine Kevin W Eliceiri, manual annotations in order to train a
learning tool for Johannes Schindelin, classifier and segment the remaining data
microscopy Albert Cardona, H automatically”. This tool provides clustering
pixel Sebastian Seung” schemes that can be customized to employ
classification” user-designed image features or classifiers.

9 Linear “Comparison of 2016 “Jean-Loup Loyer, Elsa To compare the performance of five
Regression Machine Henriques, statistical models for the prediction of
Learning MihailFontul, Steve manufacturing cost of jet engine
methods applied Wiseall” components, during the early design phase.
to the estimation The results proved “Gradient Boosted Trees”
of and “Support Vector Regression” to be very
manufacturing efficient techniques
cost of jet
engine
components”
10 Decision “Towards the 2016 “Mrinal Pandey, S. To propose integrated multiple classifiers for
Tree integration of Taruna” predicting students’ academic performance
multiple using “Decision Tree”, “K-Nearest
classifier Neighbour” and “Aggregating One-
pertaining to the Dependence Estimators (AODE)”
Student's
performance
prediction”
11 Logistic “Using Machine 2016 “Hiba Asri, Hajar To compare the performance of machine
Regression Learning Mousannif, Hassan Al learning algorithms: “Support Vector
Algorithms for Moatassime, Thomas Machine”, “Decision Tree”, “Naive Bayes”
Breast Cancer Noel” and “k Nearest Neighbors” on the Wisconsin
Risk Prediction Breast Cancer (original) datasets. The results

Published By:
Retrieval Number: I10550789S19/19©BEIESP 344 Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.I1055.0789S19 & Sciences Publication
 An Insight Into Machine Learning Techniques For Predictive Analysis And Feature Selection
and Diagnosis” show that Support Vector Machine gives the
highest accuracy of 97.13%
12 Neural “An empirical 2016 “Ruchika Malhotra” To predict defective classes using 18
Network framework for machine-learning techniques. “The proposed
defect prediction framework has been applied to seven
using machine application packages of well known, widely
learning used Android operating system viz. Contact,
techniques with MMS, Bluetooth, Email, Calendar, Gallery2
Android and Telephony”.
software”
13 Decision “Wrapper 2015 “RattanawadeePanthong, To perform optimization based on ensemble
Tree Feature Subset AnongnartSrivihok” algorithms namely “Bagging” and
Selection for “AdaBoost” by subset evaluations which are
Dimension performed using two classifiers: “Decision
Reduction Tree” and “Naïve Bayes”. “This study shows
Based on that the search technique using SFS based on
Ensemble the bagging algorithm using Decision Tree
Learning obtained better results in average accuracy
Algorithm” (89.60%) than other methods”
14 Clustering “Randomized 2015 “C. Boutsidis, A. To present feature extraction methods based
Dimensionality Zouzias, M. W. on random projections and fast approximate
Reduction fork- Mahoney and P. SVD factorizations alongwith the feature
Means Drineas” selection method for k-means clustering
Clustering”

15 Logistic “A comparative 2010 “DursunDelen” To develop an analytical model to predict

Regression analysis of and explain the reasons behind attrition of
machine fresher students
learning
techniques for
student retention
management”

16 Neural “Dropout 2009 “IoannaLykourentzou, To implement dropout prediction for e-

Network prediction in e- IoannisGiannoukos, learning courses using “feed-forward neural
learning courses Vassilis Nikolopoulos, networks”, “support vector machines” and
through the George Mpardis, Vassili “probabilistic ensemble simplified fuzzy
combination of Lo “ ARTMAP.”
machine
learning
techniques”
17 Neural “An 2000 “Carolyn Mair, To compare different machine learning
Network investigation of GadaKadoda, Martin methods to build software effort prediction
machine Lefley, Keith Phalp, systems by applying the techniques to a
learning based Chris Schofield, Martin dataset of 81 software projects. The results
prediction Shepperd, Steve proved Artificial Neural Network to be the
systems” Webster umos” most accurate technique

A. Linear Regression
III. MACHINE LEARNING TECHNIQUES Linear regression is a supervised machine learning algorithm that
is used to model a target value based on the given inputs. It is
Machine learning technique basically works by training an normally used for forecasting and finding the relation between
algorithm on a large data set, also called training set, which dependent and independent variables. If the input variable is u
has known output[11][12]. Later the algorithm is tested on a and the output variable is v, the relation between these two
test set to get the results[19][14]. Machine learning variables can be depicted using
approach can be classified as supervised learning and a linear equation
unsupervised learning. v= α1 + α2.u
The aim is to find the best value
of α1 & α2 so that we get the
Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: I10550789S19/19©BEIESP 345 & Sciences Publication
DOI: 10.35940/ijitee.I1055.0789S19
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8, Issue-9S, July 2019

best fit line. For this purpose, the cost function is the Root Mean assigned to the closest cluster centroid. This iteration goes on
Square Error(RMSE), which minimizes the difference between till no more improvements are possible and we reach the
the actual value of v and the predicted value of v. global optima.

 Hierarchical clustering: As the name suggests, this algorithm

follows a tree like structure. The clusters at one level are
formed based on the cluster in the previous level. The two
possible approaches in hierarchical clustering are:Top-down
approach and Bottom-up approach

For big data handling, k-means clustering is better than

Gradient descent is normally used to minimize RMSE. The hierarchical clustering. The time complexity of K-means is O(n)
output in case of linear regression is continuous. whereas the time complexity of hierarchical clustering is O(n2).
The results in case of hierarchical clustering can be reproduced,
B. Logistic Regression but in K-means clustering, the results may differ because of the
Logistic regression is a supervised classification algorithm that is random choice of clusters during initialization. K-means
used to classify the data according to the given input clustering performs better when the shape of clusters is circular or
features[4][18]. Unlike linear regression the output values in case spherical. The number of clusters needs to be known prior
of logistic regression is discrete. E.g. to classify whether it will whereas in hierarchical clustering, the number of clusters are
rain or not. Based on the classification category, logistic determined by interpreting the dendrogram
regression can be classified into three different types:
 Binomial classification: In Binomial classification, the D. Conjoint Analysis
class(target) variable can have two possible values like “yes”
vs “no”, “0” vs “1”, “pass” vs “fail”, “up” vs “down” etc. Conjoint analysis is used to build a model for predicting
 Multinomial classification: In multinomial classification, the customer’s preference. It is an advanced technique for
class(target) variable can have more than two values like the market research analysis in order to take insight of how the
different type of food “Chinese” vs “Italian” vs “Indian” vs customer makes complex choices. Trade-offs are required
“Asian” vs “Continental”. for making choices on daily basis, that one may not even
 Ordinal classification: In ordinal classification, the realize. Some daily life decisions such as, “what type of oil
class(target) variable has the ordered categories like feedback should I buy?” or “which area should I look for booking a
categories “Excellent” vs “Good” vs “Average” vs “Poor”. house?” all contain several elements which ultimately helps
The basic difference between linear regression and logistic in formulating a decision. Conjoint analysis is one of the
regression is that linear regression is used for prediction and most preferred models in identifying customer’s preferences
logistic regression is used for classification. Just like linear during the process of purchasing process and converting
regression, logistic regression also uses a linear equation for into a quantitative measurement. The two most important
classification purpose. The cost function used for logistic types of conjoint analysis are:
regression is called the “sigmoid function” or “logistic function”,
which returns a probability value between 0 and 1. This function  Choice-based Conjoint (CBC) Analysis: CBC analysis
maps the predicted values to probabilities is one of the most popular types of conjoint analyais
because it asks customers to imitate the purchasing
pattern as per the real market, like which products they
Unlike the straight line in linear regression, logistic regression has will choose based on certain criteria on features and
an S-shaped curve that may take a real-value number and map it
price.
into a value between 0 and 1
 Adaptive Conjoint Analysis (ACA): ACA is normally
used where the number of features/attributes are much
more than that can be handled by CBC analysis. It gives
C. Clustering
better performance for segmentation research and
Clustering is an unsupervised machine learning algorithm that is
product design, but not for the price.
used to group the data points based on their traits. The data points
in one cluster will have similar features whereas data points in
E. Neural Networks
different clusters should have different properties/features.
A neural network is a complex architecture of many layers
Clustering is normally used to extract some information from the
of interconnected units[5][19]. Each unit represents a
given data set by applying the different clustering algorithms. The
complex function of the input value. It takes input from the
two most popular algorithms used for clustering algorithms are:
previous step and gives output, which is connected to the
unit in the next step. Each such unit is called a neuron, and
 K-means clustering: It is a partitioning based iterative
the complete architecture is called a Neural Network (NN).
clustering algorithm in which the data points are partitioned
The parameters of the functions in each unit are learned as
into k clusters. The aim is to find the local maxima in each part of the training step of
iteration. The number of clusters are randomly initialized and
the neural network.
each data point is assigned to a cluster in 2D space. The
centroid of each cluster is computed and each data point is

Published By:
Retrieval Number: I10550789S19/19©BEIESP 346 Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.I1055.0789S19 & Sciences Publication
 An Insight Into Machine Learning Techniques For Predictive Analysis And Feature Selection

There is one input layer and one output layer. The design of
input and output layers is application specific. For example,
in a sentimental analysis, the input can be the sentence or
paragraphs to be analysed. The output could be a classifier,
which classifies it as neutral, negative or positive. It could
be designed for ranking (viz. from 1 to 5) on various aspects
like anger or happiness. The inner layers are called hidden
layers. In fig. 2, there is one hidden layer. The input and
hidden layer each have 4 units.

Fig. 2: Neural Network

A typical NN can be made of several hidden layers, and

several units in each hidden layer. A NN is useful because Fig.3: Decision Tree
each unit (neuron) in every layer very slightly affects the
outcome of the overall output. During the training steps of a
NN, it learns very complex features of a given problem.
The most important point in working with a decision tree is to
Each unit (neuron) in each layer can affect the overall
choose the root node, which is called as feature selection. The two
outcome only extremely slightly. But when that outcome
most popular feature selection methods are: Information gain and
passes through several layers of neurons, the effect is
Gini Index. The major advantage of decision tree is its simplicity,
compounded and gives very accurate results. The outcomes
less computation and it can handle both continuous and
of the NN are so accurate that scientists are not sure how
categorical variables.
NN are able to learn such complex features.

F. Decision Trees On the other side, Decision trees are not appropriate for
Decision tree is a supervised machine learning tool used for estimation tasks, they are prone to errors in classification
predictive modelling. It can be used for classification as well as problems where there are many classes and a small data set.
regression tasks[7]. Another name for decision trees is They can also be computationally expensive to train.
CART(Classification and Regression Trees). It is a tree like
structure that can be used to visually represent the flow of IV. FEATURE SELECTION
decision making. It is one of the simplest machine learning
algorithms. The rules in decision trees are in the form of if- With the availability of huge data sets, feature selection as a
then-else statements. The rules get complex as we go deep pre-processing step to machine learning is highly effective
in the tree and the deeper the tree, fitter the model. The in removing unimportant and redundant data, improving
nodes represent the place where we pick a feature and ask a result comprehensibility and increasing learning
question; the edges represent the answers to that question; accuracy[9][13]. Feature selection is a technique in which
and the leaf nodes represent the actual output or target the most influential features/attributes from a data set are
label. E.g. Fig. 3 shows an example of decision tree for selected for predictive analysis[1]. The irrelevant and
classifying whether a flower is Iris-Sentosa, Iris-versicolor or Iris- redundant features are eliminated from the dataset and a
Verginica based on petal width and petal length. subset of features is kept. Some supervised machine
learning algorithms have built-in feature selection like
Random Forests and Regularized Regression. Feature
selection can be supervised or unsupervised or a
combination of both[6][17]. The two most commonly used
feature selection methods are:

 Variance Thresholds: Variance thresholds eliminate

those features whose values remain almost same in
every observation (i.e. their variance falls below a
threshold). Such features
provide very little
information. E.g. in a
public health dataset, if

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: I10550789S19/19©BEIESP 347 & Sciences Publication
DOI: 10.35940/ijitee.I1055.0789S19
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-8, Issue-9S, July 2019

95% of observations are for women who are 40-year- 3. Ajmer Singh, Rajesh Bhatia, Anita Singhrova, “Taxonomy of
old, then the features 'Age' and 'Gender' can be easily machine learning algorithms in software fault prediction
eliminated without any major loss of using object oriented metrics”, Procedia Computer Science,
information[21][10]. Normalization of features is a very Volume 132, 2018, Pages 993-1001, ISSN 1877-0509,
https://doi.org/10.1016/j.procs.2018.05.115.
important step because variance is dependent on scale.
4. Arno De Caigny, Kristof Coussement, Koen W. De Bock, “A
Intuition is the base for applying variance thresholds: new hybrid classification algorithm for customer churn
features that remain same do not add much information prediction based on logistic regression and decision trees”,
to the data. It is a relatively safe and easy way for European Journal of Operational Research, Volume 269,
dimensionality reduction. It can be implemented easily Issue 2, 2018, Pages 760-772, ISSN 0377-2217,
in Python/R programming language https://doi.org/10.1016/j.ejor.2018.02.009.
5. Aggarwal D., “Sentiment Analysis: An insight into
 Correlation Thresholds: Correlation thresholds remove Techniques, Application and Challenges”, International
the highly correlated features (i.e. the features whose Journal of Computer Science and Engineering, Volume 6,
value change similarly to other's). These features do not Issue 5, 2018, pages 697-703,
add much to the information. E.g , if you had a weather https://doi.org/10.26438/ijcse/v6i5.697703
dataset with 'Temperature (Celsius)' and 'Temperature 6. RaziehSheikhpour, Mehdi Agha Sarram, Sajjad Gharaghani,
(Fahrenheit)' as different features, one of these can be Mohammad Ali ZareChahooki, “A Survey on semi-
supervised feature selection methods”, Pattern Recognition,
safely removed. The question arises which one of these
Volume 64, 2017, Pages 141-158, ISSN 0031-3203,
features should be removed. For this purpose, all pair- https://doi.org/10.1016/j.patcog.2016.11.003.
wise correlations need to be calculated. Then, if the 7. Anne-Sophie Hoffait, Michaël Schyns, “Early detection of
correlation> threshold, then the feature, that has greater university students with potential difficulties”, Decision
mean absolute correlation with rest of the features, Support Systems, Volume 101, 2017, Pages 1-11, ISSN
should be removed.Intuition is the base for applying 0167-9236, https://doi.org/10.1016/j.dss.2017.05.003.
correlation thresholds also: features that do not change 8. Chuan Liu, Wenyong Wang, Qiang Zhao, Xiaoming Shen,
much gives redundant information[22][8]. The tricky Martin Konan, “A new feature selection method based on a
part is that the threshold value needs to be set manually validity index of feature subset, Pattern Recognition Letters,
Volume 92, 2017, Pages 1-8, ISSN 0167-8655,
because if the threshold is set very low, the useful
https://doi.org/10.1016/j.patrec.2017.03.018.
information may be dropped. So, the algorithms 9. S. Sasikala, S. Appavu alias Balamurugan, S. Geetha, “Multi
with built-in feature selection method are a better Filtration Feature Selection (MFFS) to improve
choice. It can be implemented easily in Python / R discriminatory ability in clinical data set”, Applied
programming language. Computing and Informatics, Volume 12, Issue 2, 2016, Pages
117-127, ISSN 2210-8327,
V. CONCLUSION https://doi.org/10.1016/j.aci.2014.03.002.
10. Jianyu Miao, LingfengNiu, “A Survey on Feature Selection”,
Procedia Computer Science, Volume 91, 2016, Pages 919-
Predictive analysis is a process that overlaps with machine
926, ISSN 1877-0509,
learning and both go hand-in-hand. This paper tells about https://doi.org/10.1016/j.procs.2016.07.111.
the various supervised and unsupervised machine learning 11. Junfei Qiu, Qihui Wu, Guoru Ding, Yuhua Xu and
algorithms used for predictive analytics.The two most Shuo Feng, “A survey of machine learning for big data
commonly methods used for feature selection are also processing”, EURASIP Journal on Advances in Signal
discussed. Processing, 2016:67, https://doi.org/10.1186/s13634-016-
0355-x
12. Sara Landset, TaghiM. Khoshgoftaar, Aaron N. RichterEmail
ACKNOWLEDGEMENT
author and Tawfiq Hasanin, “A survey of open source tools
for machine learning with big data in the Hadoop ecosystem”,
The authors are highly grateful to the Principal, Journal of Big Data, 2015, 2:24,
Management and Department of Computer Science and https://doi.org/10.1186/s40537-015-0032-1
Engineering of JSS Academy of Technical Education, 13. RattanawadeePanthong, AnongnartSrivihok, ”Wrapper
Noida, Uttar Pradesh to provide complete support in Feature Subset Selection for Dimension Reduction Based on
carrying out the research work and writing this paper. Ensemble Learning Algorithm”, Procedia Computer Science,
Volume 72, 2015, Pages 162-169, ISSN 1877-0509,
https://doi.org/10.1016/j.procs.2015.12.117.
REFERENCES 14. Ruchika Malhotra, “A systematic review of machine learning
techniques for software fault prediction”, Applied Soft
1. Jie Cai, Jiawei Luo, Shulin Wang, Sheng Yang, “Feature Computing, Volume 27, 2015, Pages 504-518, ISSN 1568-
selection in machine learning: A new perspective”, 4946, https://doi.org/10.1016/j.asoc.2014.11.023.
Neurocomputing, Volume 300, 2018, Pages 70-79, ISSN 15. Konstantina Kourou, Themis P. Exarchos, Konstantinos P.
0925-2312, https://doi.org/10.1016/j.neucom.2017.11.077 Exarchos, Michalis V. Karamouzis, Dimitrios I. Fotiadis,
2. Concepción Burgos, María L. Campanario, David de la Peña, “Machine learning applications in cancer prognosis and
Juan A. Lara, David Lizcano, María A. Martínez, “Data prediction”, Computational and Structural Biotechnology
mining for modeling students’ performance: A tutoring action Journal, Volume 13, 2015, Pages 8-17, ISSN 2001-0370,
plan to prevent academic dropout”, Computers & Electrical
Engineering, Volume 66, 2018, Pages 541-556, ISSN 0045-
7906, https://doi.org/10.1016/j.compeleceng.2017.03.005.

Published By:
Retrieval Number: I10550789S19/19©BEIESP 348 Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.I1055.0789S19 & Sciences Publication
 An Insight Into Machine Learning Techniques For Predictive Analysis And Feature Selection

https://doi.org/10.1016/j.csbj.2014.11.005. International Journals of repute like Inderscience

16. T. Vafeiadis, K.I. Diamantaras, G. Sarigiannidis, K.Ch. and IGI Global. His research interest includes
Software Engineering, Cyber Security,
Chatzisavvas, “A comparison of machine learning techniques
Automata Theory, CBSS and ERP.
for customer churn prediction”, Simulation Modelling
Practice and Theory, Volume 55, 2015, Pages 1-9, ISSN Dr. Sonu Mittal received his Ph.D. from SGV
1569-190X, https://doi.org/10.1016/j.simpat.2015.03.003. University, Jaipur. Prior to that, he received his
17. Girish Chandrashekar, FeratSahin, “A survey on feature Master’s Degree (M. Tech.) from IGNOU. He is
selection methods”, Computers & Electrical Engineering, currently working as Associate Professor,
Department of Computer Science and
Volume 40, Issue 1, 2014, Pages 16-28, ISSN 0045-7906, Engineering at Jaipur National University,
https://doi.org/10.1016/j.compeleceng.2013.11.024. Jaipur since 2008. He has more than 14 years of
18. DursunDelen, “A comparative analysis of machine learning research and teaching experience. He has more
techniques for student retention management”, Decision than 20 publications in national/international
Support Systems, Volume 49, Issue 4, 2010, Pages 498-506, journals and conferences. His area of interest
include Machine Learning, Software
ISSN 0167-9236, https://doi.org/10.1016/j.dss.2010.06.003.
Engineering and Computer Networks.
19. IoannaLykourentzou, IoannisGiannoukos, Vassilis
Nikolopoulos, George Mpardis, VassiliLoumos, “Dropout
prediction in e-learning courses through the combination of
machine learning techniques”, Computers & Education,
Volume 53, Issue 3, 2009, Pages 950-965, ISSN 0360-1315,
https://doi.org/10.1016/j.compedu.2009.05.010.
20. Carolyn Mair, GadaKadoda, Martin Lefley, Keith Phalp,
Chris Schofield, Martin Shepperd, Steve Webster, “An
investigation of machine learning based prediction systems”,
Journal of Systems and Software, Volume 53, Issue 1, 2000,
Pages 23-29, ISSN 0164-1212,
https://doi.org/10.1016/S0164-1212(00)00005-4.
21. Witten, I.H., Frank, E., Trigg, L., Hall, M., Holmes, G.
&Cunningham, S.J. (1999), “Weka: Practical machine
learning tools and techniques with Java implementations”,
(Working paper 99/11). Hamilton, New Zealand: University
of Waikato, Department of Computer Science.
22. Hall, M. A. & Smith, “Practical feature subset selection for
machine learning”,Computer Science ’98 Proceedings of the
21st Australasian Computer Science Conference ACSC’98,
Perth, 4-6 February, 1998, Pages 181-191, Berlin: Springer.

AUTHORS PROFILE
Deepti Aggarwal has received her B. Tech.(CSE)
from MDU, Rohtak, M.Tech.(CSE) from
Rajasthan Vidyapeeth, Udaipur and pursuing
Ph.D. from Jaipur National University, Jaipur.
She has overall teaching experience of more than
18 years and currently working as Assistant
Professor at JSS Academy of Technical
Education, Noida since 2005. She has 4 papers in
international journals and conference. She has
written a book on ‘Computer Organisation’. Her
area of interest is machine learning, data mining,
operating system and compiler design.

Dr. Vikram Bali has received his B.Tech (CSE)

from REC, Kurukshetra, M.E. (CSE) from
NITTTR, Chandigarh and Ph.D from Banasthali
Vidyapith, Rajasthan. He has more than 18 years
of rich academic experience. He is a Professor &
Head of Department (CSE) at JSS Academy of
Technical Education, Noida. He is life time
member of Indian Society for Technical
Education (ISTE), Computer Society of India
(CSI) and Institution of Engineers (IE). He has
contributed 21 Research papers in International
Journal and 7 Research papers in National
Conferences/ proceedings and Edited Books. He
has also attended Faculty Enablement
programme organised by Infosys and
NASSCOM. He has been the member of board
of studies of different Indian Universities and
member of organizing committee for various
national and international seminar/conferences.
He has written books on Fundamental of "Cyber
Security and Laws", “Software Engineering" and
“Operating System”. He is reviewer to many

Published By:
Blue Eyes Intelligence Engineering
Retrieval Number: I10550789S19/19©BEIESP 349 & Sciences Publication
DOI: 10.35940/ijitee.I1055.0789S19