MITS 6002 CSU Use of Predictive Analysis in Healthcare Industry Paper Learning Outcomes The following learning outcomes have been covered in this assessme

MITS 6002 CSU Use of Predictive Analysis in Healthcare Industry Paper Learning Outcomes

The following learning outcomes have been covered in this assessment:

LO3. Conduct research on a collection of business cases and perform statistical analysis as also interpret these outcomes to recommend appropriate business directions.

LO4. Critically analyse a variety of business domains and adopt business analytics models appropriate to the domain that requires quantitative techniques for decision making.

LO5. Recommend appropriate analytic tools and techniques to resolve complex business analytics problems in various industry sectors See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/326071541
A survey of predictive analytics using big data with data mining
Article in International Journal of Bioinformatics Research and Applications · January 2018
DOI: 10.1504/IJBRA.2018.092697
CITATIONS
READS
3
5,622
2 authors:
Poornima Selvaraj
Pushpalatha Marudappa
SRM Institute of Science and Technology
SRM Institute of Science and Technology
11 PUBLICATIONS 21 CITATIONS
47 PUBLICATIONS 180 CITATIONS
SEE PROFILE
All content following this page was uploaded by Poornima Selvaraj on 30 January 2019.
The user has requested enhancement of the downloaded file.
SEE PROFILE
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/326071541
A survey of predictive analytics using big data with data mining
Article in International Journal of Bioinformatics Research and Applications · January 2018
DOI: 10.1504/IJBRA.2018.092697
CITATION
READS
1
307
2 authors:
Poornima Selvaraj
Pushpalatha Marudappa
SRM University
SRM University
8 PUBLICATIONS 4 CITATIONS
40 PUBLICATIONS 128 CITATIONS
SEE PROFILE
All content following this page was uploaded by Poornima Selvaraj on 16 July 2018.
The user has requested enhancement of the downloaded file.
SEE PROFILE
Int. J. Bioinformatics Research and Applications, Vol. 14, No. 3, 2018
A survey of predictive analytics using big data with
data mining
S. Poornima* and M. Pushpalatha
Department of CSE,
SRM University,
Chennai, Tamil Nadu, India
Email: poornima.se@ktr.srmuniv.ac.in.
Email: pushpalatha.m@ktr.srmuniv.ac.in
*Corresponding author
Abstract: Today, the world is filled with data like Oxygen. The amount of data
being harvested and eaten up is flourishing vigorously in the digital world. The
growing exploitation of novel inventions and social media leads to the
generation of huge quantities of data called Big data which can bring
remarkable information if analysed properly. Organizations may undergo for
analysis of big data to having better decisions, thus big data analytics is being
paid attention in recent times. For finding the concealed values from big data,
society requires new schemes or strategies. Predictive analytics comprises of
several statistical and analytical techniques for developing novel strategies for
the future possibilities of prediction. Therefore, Predictive analytics becomes
vital when an essential quantity of highly sensitive data has to be handled.
Based on the perceived events, future probabilities and measures are predicted.
With the aid of available data mining techniques, predictive analytics predicts
the events in future and can make recommendations called prescriptive
analytics. This review paper gives clear idea to apply data mining techniques
and predictive analytics on different medical dataset to predict various diseases
with accuracy levels, pros and cons, that concludes about the issues of those
algorithms and futuristic approaches on big data.
Keywords: big data; classification; data mining; predictive analytics.
Reference to this paper should be made as follows: Poornima, S. and
Pushpalatha, M. (2018) ‘A survey of predictive analytics using big data with
data mining’ Int. J. Bioinformatics Research and Applications, Vol. 14,
No. 3, pp.269–282.
Biographical notes: S. Poornima received her ME (CSE) Degree from Anna
University Trichy and currently she is pursuing her doing PhD (CSE) in SRM
University. She is working as an Assistant Professor in the Department of
Computer Science and Engineering at SRM University. Her research interest
includes Big Data Analytics and Data Science.
M. Pushpalatha received her PhD degree from SRM University. Currently
working as a Professor in the Department of Computer Science and
Engineering, SRM University. Her research interests include Wireless Adhoc
Networks, Distributed Systems and Wireless Sensor Networks.
Copyright © 2018 Inderscience Enterprises Ltd.
269
270
1
S. Poornima and M. Pushpalatha
Introduction
Big data is a term used for describing the exponential growth along with the structured
and unstructured availability of data. As a promising term, it contains the following
characteristics:
i
Volume: the amount of data generated.
ii
Variety: the category to which the big data belongs.
iii Velocity: the speed of generation of data.
iv
Variability: the inconsistency which can be shown by the data.
v
Veracity: accuracy corresponding to the data is dependent over the truthfulness of
the source data which are otherwise the quality of the data.
vi
Complexity: data management is becoming very complex when storing large
volumes of data from different sources.
Big data analytics is the procedure for the investigation of big data so as to reveal hidden
patterns, unknown relations and some other useful information which can be employed to
make better decisions.
Today, most of the companies store large volumes of diverse data (i.e. web logs, click
streams, sensors and several other sources). The perceptions unknown within this ‘Big
Data’ have significant business value. Several novel schemes have been developed to
handle the challenges such as volume, variety, and velocity in big data. It is
i
Apache Hadoop software that is a cost-economic, hugely scalable platform for the
analysis of big data. It can save and do the processing of petabytes of data, inclusive
of every data type which is not suitable for traditional relational database
management system (RDBMS).
ii
Not only structured query language (SQL) database lightens the restraints of the
classical RDBMS to be capable of delivering a greater performance along with
scalability. SQL databases can then have the abilities of Hadoop clusters extended by
yielding low-latency object retrieval or else other data warehouse (DW)-like
functionality.
iii Massively parallel-processing (MPP) appliances have the capacity of RDBMS-based
data warehouses extended. These systems can save and then process petabytes of
structured data.
iv
In-memory databases considerably can enhance the performance through the
elimination of most data access latencies on the shuttling of data forward and
backward between the storage systems and server processors.
A survey of predictive analytics using big data
271
In-memory databases can be considered to be an alternative in few of the MPP appliances
of today for offering realistic performance for the applications that demand high.
Predictive analytics is a type of analytics undergone on big data that deal with extracting
information from data and predict the trends and behaviour patterns. Predictive analytics
determine the possible future result of an event or even the probability of a condition that
can occur. It is one of the branches of data mining related to predict the future
possibilities and their trends. Predictive analytics is useful for analysing huge data
automatically with multiple variables; it is inclusive of decision trees, clustering, neural
nets, market basket analysis, regression modelling, hypothesis testing, decision analytics,
genetic algorithms, and text mining etc.1 It contains different view approaches like
integrated reasoning and pattern recognition along with predictive modelling. Many
researchers have interest to build an automated reasoning tool for identifying future
events and measures. Figure 1 indicates that the process of predictive analytics has to be
consistent to guarantee efficiency and accuracy of the data prediction.
Figure 1
Predictive analytics process
Predictive and prescriptive analytics is the future of data mining. The terminologies data
mining and data extraction are frequently confused with one another though the
difference is significance (Zaman). Data extraction is involved with the receipt of data
from one of the data source and having it loaded into a target database. Extraction of
data can be done in this manner, from a source system, and it is loaded in a data mart or
data warehouse. Data mining also refers extracting inconspicuous or hidden information
from data marts or data warehouses. Data mining specifies knowledge discovery as the
method used for the search of patterns in repositories of data. For knowledge discovery,
data mining employs computational strategies from statistics, machine learning, and
pattern recognition. Thus, the characteristics of data mining are described by search for
patterns hidden in the data. Various tools are developed using predictive analytical
models and strategies of data mining. The first step comprises the extraction of data by
having access to huge databases. The data obtained in this way, are then processed with
the support of sophisticated algorithms to look for concealed patterns and predictive
information. Even though statistics and data mining are related with each other, methods
employed in data mining seem to have evolved in domains except statistics.
272
S. Poornima and M. Pushpalatha
A predictive model does the analysis for identifying the patterns observed in
historical and transactional data so that different risks and potential are determined. The
forecasting models acquire the relationships between several factors to permit the
evaluation of the risks or else the opportunities that are associated with the certain listing
of conditions, thereby directing the making of decision for the candidate transactions.
Fundamental strategies for predictive analytics include
i
data profiling and transformations
ii
sequential pattern analysis
iii time series tracking 1
The first strategy includes the functions that modify the row and column attributes,
combines the fields, evaluates the dependencies, aggregates the records and data formats,
and builds rows and columns.2 Sequential pattern analysis determines that the
relationships exist between the rows in database. Sequential pattern analysis is involved
with the identification of the sequentially occurring items that are frequently seen across
the ordered transactions over time. Time series tracking can be defined as a sequence that
is ordered with values at different time intervals spaced with the equal distance.2 Time
series analysis provides the conception of data points that are plotted over time.
2
Literature survey
In past, predictive analytics can be applied in data mining for predicting future events
especially in the medical sector, business, education, and crime detection. The health
domain contains a bulk of concealed information that is significant in taking effective
decisions. Babu and Sastry (2014) concentrated over the predictive abilities of Enterprise
Resource Planning (ERP) systems, for the analysis of present data and historical facts so
that opportunities and probable risks are identified for the organisations. Analytical
decision management and business rules are utilised to make use of a decision in the form
of a service.
Bellaachia and Guven (2005) proposed predicting breast cancer lastingness using data
mining methods. The authors have examined three data mining methods such as Naïve
Bayes, propagated neural networks, and c4.5 decision tree algorithms. Naïve Bayes
method is the first method that uses the Bayesian method, because of its simple, clear,
and fast predictive nature. The second method is artificial neural networks (ANNs) that
uses multilayer network with transmission utilisation. Finally, they used c4.5 decisiontree algorithms. On the whole, the authors’ work shows that the preliminary results are
challenging prediction problem in medical data sets.
Data mining is the apt technology to predict patterns in the health sector data set.
Though it is tedious to make the prediction of few diseases such as heart attack, due to its
complexity, such tasks need more skill. Masethe and Masethe (2014) discussed to
determine heart disease using classification algorithms. Few data mining algorithms such
as j48, Naïve Bayes, REPTREE, and classification and regression trees (CART) are
applied to predict heart attacks. The author’s research work result shows that prediction
accuracy is 99%, and j48, REPTREE, and CART gave a prediction model of 89 cases
with a risk factor positive for heart attacks. From these techniques, it was identified that
prediction of diagnoses can be done by data-mining algorithms.
A survey of predictive analytics using big data
273
A medical data of large size need powerful data analysis tools for processing. Data
mining techniques can also be used for the diagnosis and predictive analysis. Ramaraj
and Thanamani (2013) proposed predictive analytics methods to identify heart diseases.
The authors’ aim was to design a predictive method for heart disease detection.
Classification accuracy report among various data mining techniques with the difference
in error rates is provided in analysis part. The authors’ final result shows that CN2Rule
performs classification more accurately than the other methods.
Nasridinov et al. (2014) discussed a study on crime pattern prediction using data
mining techniques. The authors analysed many data mining techniques with generated
test data to determine the best method to perform crime pattern prediction task.
Specifically, the authors did an extensive performance analysis of various data mining
prediction algorithms such as support vector machine (SVM), decision tree, neural
network, k-nearest neighbour, and Naïve Bayes. The authors assumed that wearable
sensor devices are attached to the clothes of the user of the proposed method. It captures
the inner temperature and heartbeat of a user and sends these data to the server to perform
emotion mining. Danger condition was identified when the user developed high
heartbeat, inner temperature, and camera surveillance that indicate the danger situation.
When a danger condition was detected, the authors employed a test data generation
method that cautiously designs test data set which comprises well-known data mining
pattern prediction algorithms. This system is useful for law enforcement and emergency
agencies to identify, analyse, and predict patterns, trends, and series, and provide useful
information to solve, reduce, and prevent various danger situations promptly.
Chandra Shekar et al. (2012) make up a better algorithm for prediction of heart
disease using case-based machine learning-based methods technique on non-binary data
sets. Mining frequent item-sets in non-binary search space presented fascinating
challenges over conventional mining in binary search space. Initially, the non-binary
search space needs innovative tactics to calculate support and must be active. As there is
a chance of removal of candidate item-set from the non-binary data set due to pruning,
applying it at a higher level may become frequent. Support calculation and candidate
generation at each level are carried out using separate mechanism. The author’s final
result was a prototype for generating frequent item sets for non-binary data set that was
developed.
Maciejewski et al. (2010) introduced a model to use in spatiotemporal data, because
the analysts are looking for the areas of space and time having unpredictably large
occurrences of events, developed a predictive visual analytics toolkit which assists the
analysts providing them with the linked spatiotemporal and statistical analytic views.
Spatiotemporal events are simulated by the system by combining the kernel density
estimation for event distribution and seasonal trend decomposition with the support of
loss smoothing for the purpose of temporal predictions. Yue et al. (2009) especially
addressed the predictive jobs which are related to the prediction of futuristic trends and
then introduced RESIN that is an artificial intelligence blackboard-based agent leveraging
the interactive visualisation and also the mixed-initiative problem solving so as to
facilitate the analysis to look for and preprocess immense quantities of data for
performing predictive analytics.
274
S. Poornima and M. Pushpalatha
Riensche et al. (2009) explained an approach for supporting the design of games in
the context of predictive analytics, developed for collecting the input knowledge,
calculating the outcomes of complicated predictive techniques and social models, and
examined those outcomes in quite an attractive manner. Huang et al. (2009) used the
predictive analytics methods for establishing a decision support system for sophisticated
network operation management and also to support the operators in predicting the
possible failures in the network and then make the network adapt as a response towards
hostile environments. The resulting decision support system facilitates the constant
monitoring of the performance of the network and converts huge quantities of data into
information that is actionable.
Sanfilippo et al. (2009) gave novel techniques for anticipatory critical thinking which
realise a multiperspective approach for performing predictive modelling in aid of
naturalistic decision making. In Banjade and Maharjan (2011), this technical work takes
the linear regression method into consideration for the analysis of large-scale data set for
providing helpful recommendations to aid the e-commerce customers using offline
computations of the outcomes of the model.
Kone and Karwan (2011) predicted the expense incurred in delivering bulk
(liquefied) gas to new customers making use of a multifactor linear regression model.
Development of a single model, i.e. evaluating all the observations one time, leads to
poor prediction outcomes. Hence, before regression analysis, a novel supervised learning
method is utilised for grouping the customers who have similarity in some or the other
perception. Hyperboxes are used to denote classes on customers, and subsequently, a
linear regression model is developed within every class. To increase with the
combination of data classification and regression, the accuracy of the prediction is
indicated.
Bhat et al. (2011) presented a new preprocessing phase along with imputation of
missing value for numerical and also categorical data. A hybrid combination consisting
of classification and regression trees (CART), genetic algorithms for imputing the
missing sequential values and self-organising feature maps (SOFM) for imputing the
categorical values are used in the work.
Bhat et al. (2009) introduced an effective imputation technique employing a hybrid
combination comprising of genetic algorithm and CART, in the role of a step of
preprocessing. The traditional neural network model is used for prediction, over the data
set that is preprocessed. Chinchor et al. (2010) address the merging of visual analytics
and multimedia analysis to tackle with the information originating from multiple sources,
having multiple aims or targets, and comprising multiple media varieties and
combinations of those types. The resultant combination results in multimedia analytics.
Razi and Athappilly (2005) performed prediction accuracy in three-way comparison that
uses nonlinear regression, CART, and NNs models employing a consistent dependent
variable along with a set consisting of dichotomous and categorical predictor variables.
Shweta et al. (2012) utilised the Naïve Bayes, decision tree, ANN, and C4.5
algorithms for the prognoses and diagnoses related to breast cancer. The results convey
that decision trees offer greater accuracy of 93.62%, ANN yields 86.5%, Naïve Bayes
produces 84.5%, and C4.5 produces 86.7%. Chaitrali et al. (2012) made use of Naïve
Bayes, decision trees, and neural network algorithms for the analysis of heart disease.
The comparison of the result shows that the Naïve Bayes attains about 90.74% of
accuracy whereas the decision trees and neural network produces corresponding 99.62%
and 100% accuracies, respectively.
A survey of predictive analytics using big data
275
Various data …
Purchase answer to see full
attachment