Chapter 6: Predictive Analytics
Introductory Caselet
Global mobile major Samsung electronics introduced a phone called Note 7 around
October 2016. Although futuristic in specifications with class-leading performance,
this phone turned out to be the darkest blot in the otherwise clean bowl of the
Samsung smartphone assembly lines. The phone had critical battery failure issues
which even resulted in a few phone explosions across the world. Airlines across the
world banned passengers from boarding the flight if they were found to be carrying
Note 7 with them. Samsung restricted the charging to 60% with a firmware upgrade
but in the end, it became a matter of so much ridicule for the company with decreased
levels of brand confidence and customers fleeing, which led the company to ultimately
recall all the phones it sold and put the lid on the project Note 7 forever – a total loss
of $18 billion.
However, rather than taking it as an incident to beat the bush around and pinning
the blame on quality control, vendors, and everyone else, Samsung took it in a positive
stride. They figured out the real issue with the battery, fixed the gaps, and exploited
the existing market sentiments cleverly by emphasizing their battery issues openly
and the steps they took to fix that goof-up and not staying behind in accepting and
recalling the defective brand like a truly professional consumer-driven company.
Results? Their competitors too had to follow the suit and declare the safety features
of their devices along with other specifications and the next phone launch of Samsung
– Galaxy S8 got rave reviews and accolades across the technical diaspora and forums.
Samsung predicted the existing anger and expected the scornful views of the loyal
base of consumers – and gave them quite a few industry-first reasons to make them
believe in their consumer-friendly image again – from issuing credit notes to
exchanging devices with S7 Edge device with extra offers to issuing apology notes and
to leading the only complete recall in the history of mobiles – Samsung won over the
market sentiments simply by predicting the outpour and anger of the customers' way
before it could get worse.
6.1 Introduction
As discussed in earlier chapters – predictive analysis is simply the analysis of the
probabilities. A scenario where the deliverables are forecasted based on the
assumptions that are arrived at after a well-calculated and thought-out scenario. For
example, a hotel chain owner might ramp down the excess offers during a restive
season of rains in a coastal area based on the predictions that there is going to be less
footfall due to the rains. Predictive analytics is complemented by effective predictive
models – models which implement the theoretical predictive analytical process.
6.2 Predictive Modelling
Predictive modeling is the process of creating, testing, and validating a model to best
predict the probability of an outcome. Many modeling methods from machine
learning, artificial intelligence, and statistics are available in predictive analytics
software solutions for this task. The model is chosen based on testing, validation, and
evaluation using the detection theory to guess the probability of an outcome in a given
set amount of input data. Models can use one or more classifiers in trying to
determine the probability of a set of data belonging to another set. The different
models available on the Modelling portfolio of predictive analytics software enable us
to derive new information about the data and to develop the predictive models. Each
model has its strengths and weakness and is best suited for types of problems.
Predictive models and analysis are typically used to forecast future probabilities.
Applied to business, predictive models are used to analyze current data and historical
facts to better understand customers, products, and partners and to identify potential
risks and opportunities for a company. It uses many techniques, including data
mining, statistical modeling, and machine learning to help analysts make future
business forecasts.
Predictive modeling is the heart and soul of business decisions.
• Building decision models is more of an art than a science.
• Creating good decision models requires:
o Solid understanding of business functional areas
o Knowledge of business practice and research
o Logical skills
• It is best to start simple and enrich models as necessary.
The biggest set of changes and advances in predictive modeling is coming about as a
result of the explosion in unstructured data—text documents, video, voice, and
images— accompanied by rapidly improving analytical techniques. In a nutshell,
predictive modeling requires structured information—the kind found in relational
databases. To make unstructured data sets useful for this kind of analysis, structured
information must be extracted from them first. One example is sentiment analysis
from Web posts. Information can be found in customer posts on forums, blogs, and
other sources that predict customer satisfaction and sales trends for new products. It
would be nearly impossible, however, to try to build a predictive model directly from
the text in the posts themselves. An extraction step is needed to get usable
information in the form of keywords, phrases, and meaning from the text in the posts,
as shown in the graphic below. Then, it’s possible to look for the correlation between
instances of the phrase “problems with the product”, for example, and spikes in
customer service calls. Predictive models are representations of the relationship
between how a member of a sample performs and some of the known characteristics
of the sample. The aim is to assess how likely a similar member from another sample
is to behave in the same manner. This model is used a lot in marketing. It helps
identify implied patterns that indicate customers’ preferences. This model can even
perform calculations at the exact time that a customer performs a transaction.
Predictive analytics techniques rely on measurable variables, manipulating metrics
to predict future behavior or outcomes given various measurable approaches.
Predictive analytics models combine multiple predictors, or measurable variables,
into a predictive model. This approach allows for the collection of data and the
subsequent formulation of a statistical model, to which additional data can be added
as it becomes available.
The addition of higher volumes of data as it becomes available creates a smart
predictive model, relying on larger and larger data sets that produce more reliable
predictions based on the volume of data analyzed. Additionally, relying on real-time
data to fuel predictive analytics models results in greater accuracy of forecasting.
A business process on Predictive Modelling
1. Creating the model: Software solutions allow you to create a model to run one or
more algorithms on the data set.
2. Testing the model: Test the model on the data set. In some scenarios, the testing is
done on past data to see how best the model predicts.
3. Validating the model: Validate the model run results using visualization tools and
business data understanding.
4. Evaluating the model: Evaluating the best-fit model from the models used and
choosing the model right fit for the data.
Predictive Modelling Process
The process involves running one or more algorithms on the data set where prediction
is going to be carried out. This is iterative processing and often involves training the
model, using multiple models on the same data set, and finally arriving at the best-
fit model based on the business data understanding.
6.2.1 Logic-driven models
Logic-driven models are created based on inferences and postulations which the
sample space and existing conditions provide. Creating logical models requires a solid
understanding of business functional areas, logical skills to evaluate the propositions
better, and knowledge of business practices and research.
To understand better, let’s take an example, a customer who visits a restaurant
around six times a year and spends around Rs. 5000/- per visit. The restaurant gets
around 40% margin on per visit billing amount.
The annual gross profit on that customer turns out to be 5000x6x0.40 = Rs. 12000/-.
30% of the customers do not return each year, while 70% do return to provide more
business to the restaurant.
Assuming the average lifetime of a customer (time for which a consumer remains a
customer) 1/.3 = 3.33 years.
So, the average gross profit for a typical customer turns out to be 12000x3.33 = Rs.
39,960/-
Armed with all the above details, we can logically conclude and can derive the
following model for the above problem statement:
Economic value of each customer (V) = (R x F x M)/D
Where R = Revenue generated per customer
F = Frequency of visits per year
M = Profit margin
D = Defection rate (Non-returning customers each year)
So, as you can see, logically driven predictive models can be derived for many
situations, conditions, problem statements, and a lot of other scenarios where
predictive analytical models provide a futuristic view based on validation, testing,
and evaluation to guess the likelihood of an outcome in a given set amount of input
data.
6.2.2 Data-Driven Models
A data-driven model is based on the data analysis of a specific system. The main data-
driven model concept is to find links between the state system variables (input and
output) without clear knowledge of the physical attributes and behavior of the
system. The data-driven predictive modeling derives the modeling method based on
the set of existing data and entails a predictive methodology to forecast future
outcomes. A company expecting losses in the current quarter due to poor market
performance and sentiments is a classic example of data-driven predictive modeling.
You have the data and you know about the data inferences. You need not predict
anything related to data, unlike Logic driven models. You are simply predicting the
outcomes based on the data. Refer to the case study in this chapter for data-driven
modeling - Samsung’s case with their product and their ensuing actions are good
examples of data-driven predictive modeling.
Self-Assessment Questions
1. Predictive analysis is all about predicting outcomes. True or False.
2. Logical Models differ from Data-driven models based on the size and type of input
variables available. True or False
Activity
Create a data-driven model using MS Excel to denote the variation in a product’s
sales for the last 3 years.
6.3 Introduction to Data Mining
Data mining is a rapidly growing field of business analytics that is focused on better
understanding characteristics and patterns among variables in large databases using
a variety of statistical and analytical tools. Many of the tools that we have studied in
previous chapters, such as data visualization, data summarization, PivotTables,
correlation and regression analysis, and other techniques, are used extensively in
data mining. However, as the amount of data has grown exponentially, many other
statistical and analytical methods have been developed to identify relationships
among variables in large data sets and understand hidden patterns that they may
contain.
Data mining can be considered part descriptive and part prescriptive analytics. In
descriptive analytics, data-mining tools help analysts to identify patterns in data.
Excel charts and PivotTables, for example, are useful tools for describing patterns
and analyzing data sets; however, they require manual intervention. Regression
analysis and forecasting models help us to predict relationships or future values of
variables of interest. As some researchers observe, “the boundaries between
prediction and description are not sharp (some of the predictive models can be
descriptive, to the degree that they are understandable, and vice versa).
In most business applications, the purpose of descriptive analytics is to help
managers predict the future or make better decisions that will impact future
performance, so we can generally state that data mining is primarily a predictive
analytic approach.
Core Ideas in Data Mining
Classification
Classification is perhaps the most basic form of data analysis. The recipient of an
offer can respond or not respond. An applicant for a loan can repay on time, repay
late, or declare bankruptcy. A credit card transaction can be normal or fraudulent. A
packet of data traveling on a network can be benign or threatening. A bus in a fleet
can be available for service or unavailable. The victim of an illness can be recovered,
still be ill, or be deceased. A common task: in data mining is to examine data where
the classification is unknown or will occur in the future, to predict what that
classification is or will be.
Similar data where the classification is known is used to develop rules, which are
then applied to the data with the unknown classification.
Prediction
Prediction is like classification, except that we are trying to predict the value of a
numerical variable (e.g., amount of purchase): rather than a class (e.g., purchaser or
non-purchaser). Of course, in classification, we are trying to predict a class, but the
term prediction in this book refers to the prediction of the value of a continuous
variable. (Sometimes in the data mining literature, the terms estimation and
regression are used to refer to the prediction of the value of a continuous variable,
and prediction may be used for both continuous and categorical data.)
Association Rules and Recommendation Systems
Large databases of customer transactions lend themselves naturally to the analysis
of associations among items purchased, or "what goes with what." Association rules,
or amity analysis, are designed to find such general. associations patterns among
items in large databases. The rules can then be used in a variety of ways. For
example, grocery stores can use such information for product placement. They can
use the rules for weekly promotional offers or for bundling products. Association rules
derived from a hospital database on patients' symptoms during consecutive
hospitalizations can help find "which symptom is followed by what other symptom"
and help predict future symptoms for returning patients. Online recommendation
systems, such as those used on [Link] and [Link], use Collaborative
Filtering, a method that uses individual users' preferences and tastes given their
historic purchase, rating, browsing, or any other measurable behavior indicative of
preference, as well as other users' histories. In contrast to association rules that
generate rules general to an entire population, collaborative filtering generates "what
goes with what" at the individual user level. Hence collaborative filtering is used in
many recommendation systems that aim to deliver personalized recommendations to
users with a wide range of preferences.
Self-Assessment Questions
1. Data Mining is the practice of scrubbing out the data from various sources for
further evaluation and analytical purposes. True or False
2. Predictive analysis deals with data mining in the same way business analytics
deals with raw data. True or False
3. Data Mining is solely a predictive analytical strategy since descriptive and
prescriptive analytics deal with data only after receiving it and predictive analysis
forecasts the data outcomes. True or False
Activity
What do Data Mining and Facebook’s friend suggestion and targeted ads have in
common? Find it out.
6.4 Data Mining Methodologies
Some common approaches in data mining include the following:
Data Exploration and Reduction- This often involves identifying groups in which
the elements of the groups are in some way similar. This approach is often used
to understand differences among customers and segment them into homogenous
groups. For example, Macy’s department stores identified four lifestyles of its
customers: “Katherine,” a traditional, classic dresser who doesn’t take a lot of
risks and likes quality; “Julie,” neotraditional and slightly more edgy but still
classic; “Erin,” a contemporary customer who loves newness and shops by brand;
and “Alex,” the fashion customer who wants only the latest and greatest (they
have male versions also).4 Such segmentation is useful in design and marketing
activities to better target product offerings. These techniques have also been used
to identify characteristics of successful employees and improve recruiting and
hiring practices.
Association - Association is the process of analyzing databases to identify natural
associations among variables and create rules for target marketing or buying
recommendations.
For example, Netflix uses the association to understand what types of movies a
customer likes and provides recommendations based on the data. [Link] also
makes recommendations based on past purchases. Supermarket loyalty cards collect
data on customers’ purchasing habits and print coupons at the point of purchase
based on what was currently bought.
Cause-and-effect modeling - Cause-and-effect modeling is the process of developing
analytic models to describe the relationship between metrics that drive business
performance—for instance, profitability, customer satisfaction, or employee
satisfaction. Understanding the drivers of performance can lead to better decisions to
improve performance. For example, the control group of Johnson Controls, Inc.,
examined the relationship between satisfaction and contract-renewal rates. They
found that 91% of contract renewals came from customers who were either satisfied
or very satisfied, and customers who were not satisfied had a much higher defection
rate. Their model predicted that a one-percentage-point increase in the overall
satisfaction score was worth $13 million in service contract renewals annually. As a
result, they identified decisions that would improve customer satisfaction. Regression
and correlation analysis are key tools for cause-and-effect modeling. Further key
methodologies have been discussed at large in the following sections:
6.4.1 Classification
Classification is the process of analyzing data to predict how to classify a new data
element. An example of classification is spam filtering in an e-mail client. By
examining the textual characteristics of a message (subject header, keywords, and so
on), the message is classified as junk or not. Classification methods can help predict
whether a credit card transaction may be fraudulent, whether a loan applicant is a
high risk, or whether a consumer will respond to an advertisement.
The classification consists of predicting a certain outcome based on a given input. To
predict the outcome, the algorithm processes a training set containing a set of
attributes and the respective outcome usually called the goal or prediction attribute.
The algorithm tries to discover relationships between the attributes that would make
it possible to predict the outcome. Next, the algorithm is given a data set not seen
before, called a prediction set, which contains the same set of attributes, except for
the prediction attribute – not yet known. The algorithm analyses the input and
produces
a prediction. The prediction accuracy defines how “good” the algorithm is. For
example, in a medical database, the training set would have relevant patient
information recorded previously, where the prediction attribute is whether or not the
patient had a heart problem. The following table illustrates the training and
prediction sets of such a database.
Among several types of knowledge representation present in the literature,
classification normally uses prediction rules to express knowledge. Prediction rules
are expressed in the form of IF-THEN rules, where the antecedent (IF part) consists
of a conjunction of conditions and the rule consequent (THEN part) predicts a certain
predictions attribute value for an item that satisfies the antecedent. Using the
example above, a rule predicting the first row in the training set may be represented
as follows:
IF (Age=65 AND Heart rate>70) OR (Age>60 AND Blood pressure>140/70) THEN Heart
problem=yes
In most cases, the prediction rule is immensely larger than the example above. The
conjunction has a nice property for classification; each condition separated by ORs
defines smaller rules that capture relations between attributes. Satisfying any of
these smaller rules means that the consequent is the prediction. Each smaller rule is
formed with AND which facilitates narrowing down relations between attributes.
How well predictions are done is measured in the percentage of predictions hit
against the total number of predictions. A decent rule ought to have a hit rate greater
than the occurrence of the prediction attribute.
Suggested Reading for References
1) Web-Site made by Christos Stergiou and Dimitrios Siganos [Link]
2) S.A. Shumsky “Selected lections about neural computing”
3) Simon Haykin “Kalman Filtering and Neural Networks” John Wiley & Sons, Inc.
ISBNs: 0-471-36998-5 (Hardback); 0-471-22154-6 (Electronic)
4) Predictive Analytics For Dummies, 2nd Edition by Anasse Bari, John Wiley &
Sons, ISBN-10: 1118728963