A REVIEW ON PREDICTIVE ANALYTICS IN DATA MINING

D.U.V.D.Prasad Teaching Asst.

Department of Mining,
Mother Teresa Institute of Science and Technology-Sathupally.
9494570424

ABSTRACT

The data mining its main process is to collect, extract and store the valuable
information and now-a-days it’s done by many enterprises actively. In advanced
analytics, Predictive analytics is the one of the branch which is mainly used to make
predictions about future events which are unknown. Predictive analytics which uses
various techniques from machine learning, statistics, data mining, modeling, and
artificial intelligence for analyzing the current data and to make predictions about
future. The two main objectives of predictive analytics are Regression and
Classification. It is composed of various analytical and statistical techniques used
for developing models which predicts the future occurrence, probabilities or events.
Predictive analytics deals with both continuous changes and discontinuous changes.
It provides a predictive score for each individual (healthcare patient, product SKU,
customer, component, machine, or other organizational unit, etc.) to determine, or
influence the organizational processes which pertain across huge numbers of
individuals, like in fraud detection, manufacturing, credit risk assessment,
marketing, and government operations including law enforcement.

KEYWORDS
Predictive analytics, Credit history, forecasting, Regression techniques

1. INTRODUCTION:
Large amount of data available in information databases becomes waste until the
useful information is extracted. Predictive analytics is the roof of advanced analytics
which is to predict the future events. Predictive analytics is capsuled with the data
collection and modelling, statistics and deployment. The figure 1 gives the basis of
predictive analysis. Based on the availability of high quality data and effective
sharing, the success of data mining relies.
 Data
Collection

Predictive

Statistics
Development

The figure 1 gives the basis of predictive analysis.

The business users allow discovering the predictive intelligence by uncovering

patterns and relationships in both the structured and unstructured data through the
data mining and text analytics along with statistics. The structured data like gender,
age, income .etc. Unstructured data are social media and they are extracted data
used in the model building process. The figure 2 explains the cycle of predictive
analytics.

Monitor Understand

the data

Deploy Prepare the

data

Evaluate Model

The figure 2 explains the cycle of predictive analytics.

2. OVERVIEW

PREDICTIVE ANALYTICS:

Predictive analytics encapsulated with statistical techniques from predictive

modeling, data mining and machine learning which are used to analyse the current
and historical facts to found the predictions about future [15]. In business, predictive
analytics are used to identify risks and opportunities. Predictive analytics are used in
various fields such as marketing, finance, travel and health care [10].

REGRESSION TECHNIQUES

A data mining function is regression which predicts a number. Regression

techniques are used to predict the age, weight, distance, and temperature [7]. In
regression task starts with a dataset in that the target values are known. Common
applications of regression are trend analysis, biomedical and financial forecasting
[4]. There were various regression algorithms which are generalized linear models
and support vector machines [13].

FORECASTING

Forecasting focus towards the predictions of the future based past and present
data [3, 14]. The two terms are focussed on the forecasting are risk and
uncertainty.

3. LITERATURE SURVEY

Carlos Márquez-Vera et al [1] A genetic programming algorithm and different

data mining approaches are proposed for solving challenge due to the high number
of factors that can affect the low performance of students and the imbalanced nature.
Methodologies used to resolve are Data Gathering, Pre-Processing, Data Mining,
and Interpretation. Interpretable Classification Rule Mining (ICRM) and SMOTE
(Synthetic Minority Over-sampling Technique) algorithms are used. Student’s data
set from where data’s are collected.WEKA tool is used. Accuracy, True positive rate,
True negative rate and Geometric mean are the parameters for performance
measurement. It results in the accurate and comprehensible classification rules and it
achieved the best predictions of student failure (98.7 %).

Branko G. Celler et al [2] The telemonitoring of vital signs from the home for the
management of patients with chronic conditions. New measurement modalities and
signal processing techniques are proposed for increasing the ions about
future.quality and value of vital signs monitoring. Automated risk stratification
algorithm is used. QRS height and width (ECG), PR (Pressure Rate), RR
(Respiratory Rate), QT (Abnormal heart rate) and Body temperature are the
parameters. jBoss tool is used. Data’s are taken from Department of Human
Services Medicare databases and Telemonitoring system data. It results in
identifying the key technical performance characteristics of at-home telemonitoring
systems.

Hao-Tsung Yang et al [3] A neural network model to predict the value of playing
style information in predicting match quality. A mix of Sternberg’s thinking style
theory and individual histories are used to categorize League of Legend (LoL)
players. The data’s are collected from LoLBase website. The algorithms used are
Feed forward/ feedback propagation algorithm and Match making algorithm. Win
rate, Match duration, Group number (ELO) are the parameters for performance
measurement. PYTHON is used. It finds that the presence of global-liberal (G-L)
style players is positively correlated with match enjoyment.

Jacob R. Scanlon et al [4] To forecast the daily level of cyber-recruitment activity

of VE (violent extremist) groups. LDA-based Topics as predictors within time series
models reduce forecast error. Latent Dirichlet allocation (LDA) algorithm is used in
forecasting. Western jihadist discussion forum is used as dataset. Number of posts
and % of recruitment post (per day) are the parameters for measurement. RTextTools
and tm text mining packages in R are used. It achieves the automatic forecast of VE
cyber recruitment using natural language processing, supervised machine learning,
and time series analysis.

Quanzeng You, Liangliang Cao et al [5] To build a reliable forecasting system for
the elections and its modelled to figure out the inter relationship between social
multimedia as image-centric and real-world entities. Competitive Vector Auto
Regression (CVAR) algorithm is used. By the competition mechanism, CVAR
compares the popularity among multiple competing candidates. Data’s are taken
from Flickr. Number of images, Number of users, images uploaded per day (IPD)
and users uploading images per day (UPD) are the parameters for measurement.
OpenCV Tool is used. As a 3 International Journal of Chaos, Control, Modelling and
Simulation (IJCCMS) Vol.5, No.1/2/3, September 2016 result CVAR is able to take
prior knowledge which leads to better performance in terms of prediction accuracy.

Sean M. Arietta et al [6] A method which found and evaluate automatically to

figure out predictive relationships between the optic presence of a city and its non-
visual attributes. Scalable distributed processing framework is implemented that
speeds up the main computational barrier by an order of magnitude. Algorithms
used are hard negative mining Algorithm, classification and recognition algorithm
and Dijkstra’s shortest path algorithm. Datasets are from 10,000 Google StreetView
panorama projections (2,000 positives, 8,000 negatives) Training dataset. The
parameters for performance measurements are Number of Detections, % of
Detections from Positive Set. MATLAB is used. As a result it’s used to define the
visual boundary of city neighbourhoods, generate walking directions that prohibit or
find exposure to city attributes and validate user-specified visual elements for
prediction.

Abish Malik, Ross Maciejewski et Al [7] A visual analytics approach that provides
decision makers with a proactive and predictive environment which helps them in
making effective resource allocation and deployment decisions. Analysts are
provided with a suite of natural scale templates and methods that enable them to
focus and drill down to appropriate geospatial and temporal resolution levels.
Prediction algorithm is used. Accuracy, Average, Count and Time are the parameters
for performance measurement. This Methodology is applied to Criminal, Traffic and
Civil (CTC) incident datasets. It provides users with a suite of natural scale
templates that support analysis at multiple spatiotemporal granularity levels.

Ronaldo C. Prati et al [8] Various graphical performance evaluation methods are

increasingly drawing the consideration of data mining. Ability to depict the trade-
offs between evaluation aspects in a multidimensional area. Graphical evaluation
methods are applicable for binary classification problems. The predictive models
deployed on Classification, Ranking, Probability estimation. The parameters for
performance measurements are True positive and false positive rate, precision and
recall. The Insurance Company (TIC).Benchmark data set includes 86- variables are
used and the tool used is WEKA- 3.5.8 version. ROC graph, ROC curve, cost lines,
cost curve Precision-recall curves, lift curve, Reliability diagram, ROI curve,
Discrimination diagram and attribute diagram are different graphical evaluation
methods. It helps on deciding the methods which is well fitted for the situations.

Gang Fang, Gaurav Pandey et al [9]Discriminative patterns can provide valuable

insights into data sets with class labels. Low-support patterns that can be discovered
using SupMaxPair. Per pattern precision, Density, dimension, count and Frequency
are the parameters for performance measurement. Frequent pattern mining
algorithm and discriminative pattern mining algorithms are used. Synthetic and
cancer gene expression data sets are used for prediction. This result in exploring
discriminative patterns by speculating patterns with relatively low support from
dense and high-dimensional data sets comparably the other approaches fall to
explore within desired amount of time.

Nanlin Jin, Peter Flach et al [10] Data mining methods for exploring incredible
consumption patterns and their associated descriptive models from smart electricity
meter data. Target concept, Target type, Double regression, Coverage and Strategy
type are the parameters for performance measurement. Subgroup discovery
algorithms and S-Transform algorithms are used. Data used were collected by the
Energy Demand Research Project (EDRP). Cortana tool is used. This approach
 outperforms more conventional data mining methods in terms of their predictive
power and classification accuracy, while consuming similar computational resource.

4. COMPARISONS ON DIFFERENT PREDICTIVE ANALYTIC TECHNIQUES

Title Techniques Datasets Parameter Conclusion

And
Algorithms
Predicting Student Data Gathering, Pre- Student’s Accuracy, True Accurate And
Failure At School Processing, Data Data Set Positive Rate, Comprehensib
Using Genetic Mining, And True Negative le
Programming Interpretation. Rate And Classification
And Different Interpretable Geometric Mean Rules
Data Mining Classification Rule
Approaches With Mining (ICRM) And
High Dimensional SMOTE (Synthetic
And Imbalanced Minority Over-
Data Sampling Technique)
Algorithms
Home Automated Risk Departme QRS Height And Identifying The
Telemonitoring Stratification nt Of Width (ECG), Key Technical
Of Vital Signs Algorithm Human PR (Pressure Performance
Technical Services Rate), RR Characteristics
Challenges And Medicare (Respiratory Of At-Home
Future Directions Databases Rate), QT Telemonitoring
And (Abnormal Heart Systems.
Telemoni Rate) And Body
toring Temperature
System
Data

Thinking Style Feed Forward/ Lolbase Win Rate, Match Finds That
And Team Feedback Propagation Website Duration, Group The Presence
Competition Algorithm And Match Number (ELO) Of Global-
Game Making Algorithm. Liberal (G-L)
Performance And Style Players
Enjoyment Is Positively
Correlated
With Match
Enjoyment

Forecasting Latent Dirichlet Western Number Of Posts Achieves The

Violent Extremist Allocation (LDA) Jihadist And % Of Automatic
Cyber Algorithm Discussio Recruitment Post Forecast Of
Recruitment n Forum (Per Day) VE Cyber
Recruitment
Using Natural
Language
Processing,
Supervised
Machine
Learning, And
Time Series
Analysis
A Multifaceted Competitive Vector Flickr Number Of Better
Approach To Auto Regression Images, Number Performance In
Social (CVAR) Algorithm Of Users, Images Terms Of
 Based Prediction Day (IPD) And Accuracy.
Of Elections Users Uploading
Images Per Day
(UPD)
City Forensics: Hard Negative Mining Google Number Of Define The
Using Visual Algorithm, Streetvie Detections, % Of Visual
Elements To Classification And w Detections From Boundary Of
Predict Non- Recognition Panorama Positive Set City
Visual City Algorithm And Projectio Neighbourhoo
Attributes Dijkstra’s Shortest ns ds, Generate
Path Algorithm Training Walking
Directions
That Prohibit
Or Find
Exposure To
City Attributes
And Validate
User-Specified
Visual
Elements For
Prediction
Proactive Prediction Algorithm Criminal, Accuracy, It Provides
Spatiotemporal Traffic And Average, Count Users With A
Resource Civil And Time Suite Of
Allocation And (CTC) Natural Scale
Predictive Visual Incident Templates That
Analytics For Datasets. Support
Community Analysis At
Policing And Law Multiple
Enforcement Spatiotemporal
Granularity
Levels.

A Survey On Graphical Evaluation Insurance True Positive It Helps On

Graphical Methods Methods Company And False Deciding The
For Classification (TIC).Be Positive Rate, Methods Which
Predictive nchmark Precision And Is Well Fitted
Performance Data Set Recall For The
Evaluation Situations.

Mining Low- Frequent Pattern Synthetic Per Pattern Exploring

Support Mining Algorithm And Precision, Discriminative
Discriminative And Discriminative Cancer Density, Patterns By
Patterns From Pattern Mining Gene Dimension, Speculating
Dense And Algorithms Expressio Count And Patterns With
High- n Data Frequency Relatively Low
Dimensional Sets Support From
Data Dense And
High-
Dimensional
Data Sets
Comparably
The Other
 Approaches
Fall To Explore
Within Desired
Amount Of
Time.

Subgroup Subgroup Discovery Energy Target Concept, Outperforms

Discovery In Algorithms And S- Demand Target Type, More
Smart Electricity Transform Algorithms Research Double Conventional
Meter Data Project Regression, Data Mining
(EDRP) Coverage And Methods In
Strategy Type Terms Of Their
Predictive
Power And
Classification
Accuracy,
While
Consuming
Similar
Computational
Resource.

5. CONCLUSION

Predictive analytics is the future of data mining .This study focus towards the
predictive analytics, regression techniques and forecasting in knowledge discovery
domain. Business intelligence is used in predictive analytics for modelling and
forecasting. Predictive analytics are more efficient in choosing marketing methods
and helpful in social media analytics.

REFERENCES

[1] Carlos Márquez-Vera, Alberto Cano, Cristóbal Romero, Sebastián

Ventura,“Predicting student failure at school using genetic programming and
different data mining approaches with high dimensional and imbalanced data”,
Springer Science+Business Media, LLC 2012.

[2] Branko G. Celler, and Ross S. Sparks, “Home Telemonitoring of Vital Signs
Technical Challenges and Future Directions”, IEEE Journal Of Biomedical And
Health Informatics, Vol. 19, No. 1, January 2015.

[3] Hao Wang, Hao-Tsung Yang, and Chuen-Tsai Sun,” Thinking Style and Team
Competition Game Performance and Enjoyment”, IEEE Transactions On
Computational Intelligence And Ai In Games, Vol. 7, No. 3, September 2015.
[4] Jacob R. Scanlon and Matthew S. Gerber, “Forecasting Violent Extremist Cyber
Recruitment”, IEEE Transactions On Information Forensics And Security, Vol.
10, No. 11, November 2015.

[5] Quanzeng You, Liangliang Cao, Yang Cong, Xianchao Zhang, and Jiebo Luo” A
Multifaceted Approach to Social Multimedia-Based Prediction of Elections”,
IEEE Transactions On Multimedia, Vol. 17, No. 12, December 2015.

[6] Sean M. Arietta Alexei A. Efros Ravi Ramamoorthi Maneesh Agrawala, “City
Forensics: Using Visual Elements to Predict Non-Visual City Attributes”, IEEE
Transactions On Visualization And Computer Graphics, Vol. 20, No. 12,
December 2014.

[7] Abish Malik, Ross Maciejewski, Sherry Towers, Sean McCullough, and David
S. Ebert,” Proactive Spatiotemporal Resource Allocation and Predictive Visual
Analytics for Community Policing and Law Enforcement”, IEEE Transactions
On Visualization And Computer Graphics, Vol. 20, No. 12, December 2014.

[8] Ronaldo C. Prati, Gustavo E.A.P.A. Batista, and Maria Carolina Monard,” A
Survey on Graphical Methods for Classification Predictive Performance
Evaluation”, IEEE Transactions On Knowledge And International Journal of
Chaos, Control, Modelling and Simulation (IJCCMS) Vol.5, No.1/2/3,
September 2016 Data Engineering, Vol. 23, No. 11, November 2011.

[9] Gang Fang, Gaurav Pandey, Wen Wang, Manish Gupta, Michael Steinbach, and Vipin
Kumar,” Mining

Low-Support Discriminative Patterns from Dense and High-Dimensional

Data”, IEEE Transactions On Knowledge And Data Engineering, Vol. 24, No.
2, February 2012 .

[10]Nanlin Jin, Peter Flach, Tom Wilcox, Royston Sellman, Joshua Thumim, and
Arno Knobbe,” Subgroup Discovery in Smart Electricity Meter Data”, IEEE
Transactions On Industrial Informatics, Vol. 10, No. 2, May 2014.

[11] Hao Wang, Hao-Tsung Yang, and Chuen-Tsai Sun,” Thinking Style and Team
Competition Game Performance and Enjoyment”, IEEE Transactions On
Computational Intelligence And Ai In Games, Vol. 7, No. 3, September 2015.

[12]Quanzeng You, Liangliang Cao, Yang Cong, Senior Member, IEEE, Xianchao
Zhang, and Jiebo Luo, Fellow, IEEE.” A Multifaceted Approach to Social
 Multimedia-Based Prediction of Elections”, IEEE Transactions On Multimedia,
Vol. 17, No. 12, December 2015.

[13]Yun Wang and Sudha Ram,” Predicting Location- Based Sequential Purchasing
Events by Using Spatial, Temporal, and Social Patterns”, IEEE Intelligent
Systems, May/June 2015.

[14]Jesse Rio Russell,” Predictive analytics and child protection: Constraints and
Opportunities”, Child Abuse & Neglect 46 (2015) 182–189- ELSEVIER.

[15]Karel Dejaeger, Wouter Verbeke, David Martens, and Bart Baesens,” Data
Mining Techniques for Software Effort Estimation: A Comparative Study”,
IEEE Transactions On Software Engineering, Vol. 38, No. 2, March/April 2012.

[16]Leonardo Feltrin,” KNIME an Open Source Solution for Predictive Analytics in

the Geosciences”, IEEE Geoscience and remote sensing magazine, December
2015.

[17]Josep Ll. Berral, Nicolas Poggi, David Carrera, Aaron Call, Rob Reinauer,
Daron Green,” ALOJA: A Framework for Benchmarking and Predictive
Analytics in Big Data Deployments”, IEEE Transactions on Emerging Topics in
Computing • November 2015.

[18]Minghui Zhou and Audris Mockus,” Who Will Stay in the FLOSS Community?
Modeling Participant’s Initial Behavior”, IEEE Transactions On Software
Engineering, Vol. 41, No. 1, January 2015 .

[19]Sean M. Arietta Alexei A. Efros Ravi Ramamoorthi Maneesh Agrawala, “City

Forensics: Using Visual Elements to Predict Non-Visual City Attributes”, IEEE
Transactions On Visualization And Computer Graphics, Vol. 20, No. 12,
December 2014.

[20] Francisco C. Pereira, Member, IEEE, Filipe Rodrigues, Evgheni Polisciuc, and Moshe Ben-Akiva”,
Why so many people? Explaining Nonhabitual Transport Overcrowding With Internet Data”,
IEEE Transactions On Intelligent Transportation Systems, Vol. 16, No. 3, June 2015.