Predictive Analytics for Predicting Customer

Behavior
Kridanto Surendro
Asniar
School of Electrical Engineering and Informatics
School of Electrical Engineering and Informatics
Institut Teknologi Bandung
Institut Teknologi Bandung
Bandung, Indonesia
Bandung, Indonesia
[email protected]
[email protected]

Abstract—The development of the internet has caused Predictive analytics is used to predict trends, improve
digitalization of data which opens up big data opportunities. performance, encourage decision making, and predict
Digital data in large numbers leaves traces of what customers behavior [4].
see, what they read, their involvement and behavior, judgment,
about their interests and preferences so as to provide a large Predictive analytics is also defined as the use of statistical
amount of data that can be mined for learning experiences. or machine learning methods to make predictions about
The big data value lies in the results of analysis and predictions future or unknown results [5]. Predictive analytics deals with
or actions taken from the results of the analysis and prediction. the analysis of historical data to provide predictions about
Predictive analytics is data utilization, statistical algorithms, future events. For predictions, predictive analytics applies
and machine-learning techniques to identify possible trends, several data mining techniques and statistics, such as
events, and behaviors in the future based on historical data. clustering, classification, regression, etc. By applying this
This paper tries to propose predictive analytics to predict technique, predictive analytics builds a predictive model that
customer behavior by using behavior informatics and analytics represents certain conditions between available features or
approach so that deeper insight into customer behavior can be predictors associated with the event [6]. Predictive analytics
obtained to support predictive analysis in order to improve attempts more on implicit analysis that focuses on predictive
business decision making. modeling which refers to the type of data analysis that makes
predictions about unknown future events and reveals the
Keywords— predicitive analytics, predictive model, customer
behavior, behavior informatics and analytics, behavioral data,
reasons behind them, usually with further analysis [7].
behavior analysis. Predictive analytics can bring clarity and consistency to
situations about possible future behavior or uncertain
conditions [8].
I. INTRODUCTION
Extensive investment in the infrastructure business has A. Predictive Analytics Life Cycle
increased the ability to collect data throughout the company. The success of predictive analytics is more about better
Almost every aspect of the business is open for data decision making. Previously, with a low data volume,
collections: operational, manufacturing, supply chain intuitive decision making was still successful. But because
management, customer behavior, marketing campaign the size of the data has reached extraordinary proportions has
performance, and workflow procedures. Information on the increased, the human ability to make intuitive decisions fully
availability of external data is also increasingly widespread, has been reduced. As a result, decision making based on data
such as marketing trends, industry news and competitor becomes more interpreted as ensuring a reasonable pathway
movements. Computer technology has become much for better decision making. Data driven decision making is
stronger, internet networks are everywhere, and algorithms often based on quantitative models that are made using
have been developed to be able to connect datasets to enable closed-loop processes, commonly referred to as cycles, as in
a wider and deeper analysis than before. The availability of Figure 1 [9].
this vast data has led to an increasing interest in methods for
extracting useful information and knowledge from data. This 1. Problem
Identificati
leads to the use of data for competitive advantage [1]. on and
Definition

Today, different technologies such as cloud computing, 7 Feed

Back
2. Design
and Built
SOA, jointly change information technology which in turn
creates new complexity in the calculation of data. For this
reason, fast and dynamic data analysis is needed for both
structured and unstructured data [2]. Big data without 6.
3. Data
analytic is just a lot of data. The aim and goal of analytic is Actionable
Analytics
Acquisition

to improve business decisions by revolutionizing the way to

analyze and utilize data [3].
4.
Exploratio
5. Analysis
n and
II. PREDICTIVE ANALYTICS Reporting

Predictive analytics is the use of data, statistical Fig. 1. Predictive Analytics Life Cycle [9]
algorithms, and machine-learning techniques to identify Figure 1 represents the predictive analytics life cycle. It
possible results in the future based on historical data. can be seen that the life cycle begins with problem
Predictive models use known results to develop models that identification and definition then continues with the design
can be used to predict values for different data or new data.

230
978-1-5386-8448-1/19/$31.00 ©2019 IEEE
and build of an analytical framework (if none are available). objective data and failure to consider micro-level decisions
After that proceed to the data management, reporting, and and actions that collectively encourage macro-level behavior
visualization stages. For further, analysis is carried out to [13]. Descriptive variables are not available or inadequate to
produce models, execution and testing. The cycle ends with distinguish differences and similarities between customers
feedback which then becomes another input for the next [14]. For this reason, a new framework is needed to
predictive analytics project. overcome this by advocating objective and perceptual
information and describing behavior into a series of stages
III. PREDICTING CUSTOMER BEHAVIOR that are closely integrated as represented by Figure 2 [5].
Regarding customer behavior, predictive analtytic is a
general term for the application of mathematical techniques
to predict customer behavior. The result is usually a score or
code for each customer, which indicates the possibility of
their behavior in the future. For example, a score can
represent the possibility of them buying another product
from a company. This score or code can also place customers
into groups that might have different needs for management
and communication, or for products and services of a
company [10]. In practice, the data set used to predict
increases every day. Like on some shopping websites, users
do different things every day. If you want to use behavioral
data to predict a customer's future behavior, behavior data
must be added to the existing data set [11]. Customer
behavior analysis can help service providers to improve the Fig. 2. Behavior Prediction Framework [5]
quality of their services in order to increase customer
satisfaction [12]. Figure 2 illustrates the framework of prediction behavior.
This framework outlines complex behavior predictive tasks
Especially for marketing purposes, predictive analytic
into sequences of interrelated prediction stages that take into
tries to predict future behavior of customers, such as their account perceptual and objective variables. Predictive
responses to promotional activities from very incomplete models that work better (especially certain perceptual
information where the marketing database usually contains predictors and their interactions with the decision stage) can
little information about customer demographics, product enhance our understanding of behavioral phenomena in
needs, and interest in buying. Even the marketing data base various application domains.
often has no information about products that customers buy
from other companies [10]. 'Behavior' implies data about what customers do. This
covers a very wide area and produces the largest number of
TABLE I. PREDICTIVE ANALYTICS FOR MARKETING [10] data items, for example on purchase transaction product
usage. Online behavior also shows customer interests, such
Marketing Application Business Question as web browsing and the use of social media such as
Customer recruitment Which customers are most likely to buy
from a prospect database products/services?
Facebook and Twitter. The behavior variable is ideal for
Cross-sell/up-sell Which customers of Product X are most likely predictive analysis, because it should be more accurate and
campaign to buy Product Y? Which customers of available for all customers of that product or service.
products Z are most likely to buy more Behavior variables can tell 'what the customer is doing' [10].
products Z?
Next-best offer What products/services are each customer Customer activity can change rapidly in response to new
likely buy next? needs and changes, therefore behavioral data sometimes can
Customer retention Which customers are most likely not to buy be more predictive than demographics. Behavioral data also
again? can track behavior patterns where the first indication of
Customer lifecycle How long before each customer becomes
management likely to lapse or attrite?
change can be obtained. For example, if a customer has
Win-back campaign Which past customers are most likely to purchased a new cell phone on a different network, there will
respond to a win-back offer? be a sharp reduction in calls made on the previous network.
Customer future value or What is the predicted future value of Similarly, if a new supermarket is opened in the city, then
lifetime value purchasing, or contributions to profit, for each this will have a clear impact on the level and value of
customer? transactions with existing stores. At a more refined level,
behavioral data can be an indication of a more significant
Table 1 shows further marketing applications of event.
predictive analytics along with business questions that can be
addressed. It is seen that the most common application of
IV. BEHAVIOR INFORMATICS AND ANALYTICS
predictive analytics is to predict customer behavior.
Technological developments have enabled the collection
One challenge that arises in the area of research on of data on human behavior in both rural and urban areas [15].
predictive analytics is behavior prediction, namely the ability The value of understanding human behavior is undoubted
to accurately predict individual decisions and actions from and important in almost all application scenarios related to
predictive analytics. Predicting behavior can also help inform humans, from commercial services, community management
behavioral theory. Behavior prediction is increasingly to public security [16]. In recent years, there have been many
important, but the existing traditional modeling approach still attempts to represent human interactive behavior in the form
has drawbacks, namely high dependence on observed of computing [17].

231
 Behavior Informatics and Analytics is a scientific field as dealing with the interaction of customer-officer
that aims to develop methodologies, techniques and practical interaction, counterterrorism and online community
tools for representing, modeling, analyzing, understanding monitoring [18].
and/or utilizing symbolic behavior and/or mapped behavior,
BIA appears as an important tool for finding behavior
behavioral interaction and network, behavior patterns,
behavior impacts, the formation of behavior-oriented groups intelligence and behavior insight. As a computational
concept, behavior captures the demographic aspects of
and collective intelligence, and behavioral intelligence
emergence. In essence, Behavior Informatics and Analytics subject and behavioral object; social relationship or norms
that regulate the interaction between individual or group
strive to bring computing technology and tools to deeply
behavior; sequence of behaviors or networks and dynamics;
understand behavior and social behavior network [18].
and the impact or effects produced by the behavior carried
Behavior is increasingly recognized as a key component out by the subject on the object. BIA analyzes behavior
in business intelligence and problem solving. Different from patterns and their impact in terms of developing methods of
traditional behavior analysis, which focuses primarily on mining behavior patterns, presenting behavior patterns, and
implicit behavior and explicit business appearance as a result transforming them into business rules for decision support.
of business use and customer demographics, the Behavior BIA results consist of behavior patterns and business rules
Informatics approach supports explicit and quantitative that are more appropriate for business decision making
behavioral involvement through the conversion of
transactional data to behavioral data, for further analysis of V. PREDICTIVE ANALYTICS WITH BEHAVIOR INFORMATICS
the original patterns and impact of behaviour [19]. AND ANALYTICS APPROACH
Construction of this behavior is to change transactional One challenge that arises in the area of research on
data or data collected and managed by existing data predictive analytics is behavior prediction, namely the ability
management systems into behavioral data in accordance with to accurately predict individual decisions and actions from
certain behavioral models, logic, relationships, and semantics predictive analytics [5]. In recent years, there have been
that can be collected from domain knowledge and domain many attempts to represent human interactive behavior in the
intelligence. After behavioral data is formed, behavioral form of computing [17].
insights in behavioral data can be analyzed and extracted to
find behavioral intelligence that reflects behavioral insights For this reason, this study tries to propose Predictive
that are consistent with behavior in the physical world [20]. Analytics for Predicting Customer Behavior with Behavior
Informatics and Analytics Approach as shown in Figure 5
below.

1
Behavior
Preprocessing
Analysis with
Historical with Behavioral Behavioral
Predictive Model
Data Data Data
Modelling
Construction
Algorithm

New Data Model Predictions

Fig. 3. Behavioral Data [18]

Based on Figure 3 above, it can be seen that from a

conceptual perspective, BIA (Behavior Informatics and Fig. 5. Predictive Analytics for Predicting Customer Behavior
Analytics) is a process that transforms transactional data- At the preprocessing stage, with the behavioral data
oriented entity relationships into behavioral feature-oriented construction, historical data containing normal transaction
data through behavior modeling. data will be converted into behavior-oriented feature space
Behavior Behavior containing behavioral data where the elements of customer
Behavioral Pattern Analysis behavior become the main proportion of the data set.
Modeling Data Mapping transactional data into behavioral data is the main
Behavior
Transactional Data thing at this stage.
Impact Analysis
By using predictive modeling algorithm, behavior pattern
analysis will identify patterns of customer behavior and then
Behavior build predictive models based on behavioral patterns
Business
Business Application
Rules
Presentation
founded from behavioral data. Predictive modeling is the
process by which models are made to predict results [21].
Data mining and machine learning techniques, such as:
Fig. 4. Proses BIA [18]
Classsification, Regression, Clustering, and Association
Figure 4 shows the BIA Process. BIA consists of the Rules are used in this predictive modeling. Furthermore, the
main components including behavior modeling and predictive model produced will be used to predict customer
representation, behavioral data construction, behavior impact behaviour in the future.
modeling, behavior pattern analysis, and behavior
presentation. BIA can greatly complement existing facilities Classification is used if the predicted results are in the
for a more informative combination of social patterns and form of categories while regression is used if the prediction
solutions for solving critical problems in various fields such results are in numerical form. Descriptive modeling or

232
clustering is used to assign observations to the cluster so that ACKNOWLEDGMENT
observations in the same cluster are similar. Meanwhile This work is in part of our PhD research. Thanks are
association rules can be used to find interesting associations given to Telkom University for supporting this PhD
between observations [21]. Education Funding.
A. Classification
REFERENCES
Mining data using classification is intended to predict the
[1] F. Provost and T. Fawcett, Data Science for Business. 2013.
value of categorical variables (target or class). The model is [2] P. Wazurkar and R. S. Bhadoria, “Predictive Analytics in Data
built based on one or more numerical variables and/or Science for Business Intelligence Solutions,” in International
categorical variables (predictors or attributes). Conference on Communication Systems and Network
Technologies, 2017, pp. 367–370.
[3] D. Nauck, “Predictive analytics and proactive service,” in IET
B. Regression
Seminar on Data Analytics 2013: Deriving Intelligence and
Mining data using regression is intended to predict target Value from Big Data, 2013, pp. 4–4.
values in the form of numerical variables by constructing [4] S. Banumathi and A. Aloysius, “PREDICTIVE ANALYTICS
models based on one or more predictors (numerical variables CONCEPTS IN BIG DATA- A SURVEY,” Int. J. Adv. Res.
Comput. Sci., vol. 8, no. 0976, pp. 27–30, 2017.
and categorical variables). [5] A. Abbasi, R. Y. K. Lau, and D. E. Brown, “Predicting
behavior,” IEEE Intell. Syst., vol. 30, no. 3, pp. 35–43, 2015.
C. Clustering [6] L. J. Fülöp, Á. Beszédes, G. Tóth, H. Demeter, L. Vidács, and L.
Farkas, “Predictive Complex Event Processing: A Conceptual
A cluster is a subset of data which are similar. Clustering Framework for Combining Complex Event Processing and
is also called unsupervised learning. The process divides a Predictive Analytics,” in Proceedings of the Fifth Balkan
dataset into groups such that the members of each group are Conference in Informatics (BCI ’12), 2012, pp. 26–31.
as similar (close) as possible to one another, and different [7] L. Cao, “Data Science: A Comprehensive Overview
groups are as dissimilar (far) as possible from one another. LONGBING,” ACM Comput. Surv., vol. 50, no. 3, 2017.
[8] A. H. Yousef, A. F. Fahmy, and H. K. Mohamed, “On the use of
predictive analytics techniques for network elements failure
D. Association Rules prediction in telecom operators,” in ICENCO 2017 - 13th
Association Rules is used to find that all sets of items International Computer Engineering Conference: Boundless
Smart Societies, 2018, vol. 2018–Janua, pp. 250–255.
(itemsets) that have support greater than the minimum [9] A. Cordoba, Understanding the Predictive Analytics Life Cycle.
support and then using the large itemsets to generate the Hoboken, New Jersey: John Wiley & Sons, Inc., 2014.
desired rules that have confidence greater than the minimum [10] B. Leventhal, Predictive Analytics for Marketers : Using Data
confidence. The lift of a rule is the ratio of the observed Mining for Business Advantage. Kogan Page Limited, 2018.
support to that expected if X and Y were independent. [11] B. W. Shang and K. Wang, “A Data Flow Model to Solve the
Data Distribution Changing Problem in Machine Learning,” in
Market basket analysis is a typical and widely used example 3rd Annual International Conference on Information Technology
of association rules. and Applications (Ita 2016), 2016, vol. 7.
[12] K. Rattanathavorn and W. Premchaiswadi, “Analysis of customer
By utilization behavioral data to be modeled and further behavior in a call center using fuzzy miner,” in International
analyzed, it is expected that deeper insight into customer Conference on ICT and Knowledge Engineering, 2015, vol.
behavior can be obtained to support more accurate prediction 2015–Decem, pp. 137–141.
analysis in order to increase business decision making. [13] D. E. Brown, A. Abbasi, and R. Y. K. Lau, “Predictive analytics:
Predictive modeling at the micro level,” IEEE Intell. Syst., vol.
30, no. 3, pp. 6–8, 2015.
VI. CONCLUSION [14] P. W. Murray, B. Agard, and M. A. Barajas, “Market
segmentation through data mining: A method to extract behaviors
Predictive analytics is used to predict trends, events and from a noisy data set,” Comput. Ind. Eng., vol. 109, pp. 233–252,
behavior in the future based on data which is intended to 2017.
support improvement data driven decision making. [15] R. Punj and R. Kumar, “Technological aspects of WBANs for
Predictive analytics is used more for marketing purposes, health monitoring: a comprehensive review,” Wirel. Networks,
especially to predict customer behavior where companies vol. 3, pp. 1–33, 2018.
start recording customer responses and transactions, then use [16] P. Cui et al., “Uncovering and Predicting Human Behaviors,”
IEEE Intell. Syst., vol. 31, no. 2, pp. 77–88, 2016.
data to target their marketing campaigns. Customer activity [17] M. Kouzehgar, M. Badamchizadeh, and M. R. Feizi-Derakhshi,
can change rapidly in response to new needs and changes in “Ant-Inspired Fuzzily Deceptive Robots,” IEEE Trans. Fuzzy
life. Therefore, behavioral data can sometimes be more Syst., vol. 24, no. 2, pp. 374–387, 2016.
predictive than demographic data. [18] L. Cao, “Behavior Informatics and Analytics: Let behavior talk,”
in Proceedings - IEEE International Conference on Data Mining
With the behaviour informatics and analytics approach, Workshops, ICDM Workshops 2008, 2008, pp. 87–96.
in which there is behavioral data construction that maps [19] L. Cao, “In-depth behavior understanding and use: The behavior
transaction data into behavioral data and behavioral analysis informatics approach,” Inf. Sci. (Ny)., vol. 180, no. 17, pp. 3067–
3085, 2010.
that identifies behavioral patterns, it is expected to be gained [20] L. Cao, Data Science Thinking. Springer, 2018.
a deeper understanding of customer behavior to support more [21] M. Ouahilal, M. El Mohajir, M. Chahhou, and B. E. El Mohajir,
accurate prediction analysis to increase business decision “A comparative study of predictive algorithms for business
making. Future research is a case study using a predictive analytics and decision support systems: Finance as a case study,”
modeling algorithm method to predict customer behavior in 2016 International Conference on Information Technology for
Organizations Development, IT4OD 2016, 2016, pp. 1–6.
with a behavioral analysis approach.

233