Business analytics and data mining

Overview of BA Discussion
 Business Analytics (BA)
 Overview
 History
 Types of Business Analytics
 Real world examples
 Challenges
 Relations to Data Mining

Business Analytics (BA) : an
overview
 BA can be considered a subset of Business intelligence
 A set of skills, technologies, applications and practices
 exploration and investigation of past business performance
to gain insight and drive business planning.
 Like Business Intelligence, BA can focus either on the
business as a whole or only on segments of it
 Focuses on developing new insights and understanding
of performance based on data and statistical methods

BA : Short History
 Analytics in business dates far before computing
 Frederick Taylor, father of scientific management, 19th
century
 time management exercises used in industrial settings
 Henry Ford : assembly line pacing used to improve output
and business profitability
 BA becomes widespread when computers were used in
DSS systems in the 60’s
 Evolved into ERP, data warehouses, etc.

Types of Business Analytics
 Reporting or Descriptive Analytics
 Affinity grouping
 Clustering
 Modeling or Predictive analytics

BA: Reporting
 Based on the need to locate and distribute business
insights and experiences
 Often involves ETL procedures used alongside a data
warehousing scheme
 The data is then collected, quantified, and organized
using reporting tools
 Reporting, allows for information describing different
views of an enterprise to come together one place
 A user could query a production and marketing database to
determine if production of a product could be moved closer
to where a product is sold

BA: Affinity grouping
 A tool used by businesses and
organizations to take ideas
and data and organize them.
 Often takes the form of an affinity diagram
 Enables data and ideas stemming from
brainstorming to be sorted into groups
 Sorting is based on their natural relationships

BA: Clustering
 Placing a set of objects into groups (called clusters) so
that the objects in the same cluster are more similar (in
some sense or another) to each other than to those in
other clusters – wikipedia
 Is a main task of explorative data mining and statistical
data analysis
 Clustering is a general task that does not have one set
solution
 Clustering can be hard or fuzzy
 Can be done by people or machines
 The latter is preferred

BA: how do we model clusters?
 Connectivity models – how data can be connected to
other points
 Density models – defining a cluster by determining where
sets of data points are densest
 Distribution model – clusters are modeled using statistical
distributions
 Expectation maximization

BA: Predictive Analysis
 Stems from the desire to predict future events through
analyzing data an enterprise has collected
 Pattern exploitation results in the identification of
opportunities and also risks
 Allow relationships in disparate data to be identified
 Helps guide in decision making in a business
 Is often implemented in the form of data mining

BA : Examples
 Credit company– uses business analytics to track credit risk of
customers as well as matching customers to offerings
 Sales and offers – companies can track customer interaction,
and use that information to determine appropriate product
offerings.
 Sales groups can use BA to optimize inventory and analyze
past sales
 Could measure peak purchasing times for products
 Could decide whether or not to stock poorly selling items
 Give examples of business cases where data mining might be
useful, and describe how data mining would be used
 Preventing credit card fraud through detecting spending patterns
 Inventory management by tracking sales

BA : Challenges
 Acquiring sufficient volumes of high quality data
 Most data acquired in the field is unsorted and appears in
many different formats
 When dealing with high volume data, deciding what is
important and what is noise
 Rapidly reacting storage structures
 BA can influence customer interactions, and as such that
information must be available fast
 Ex: a customized sales pitch

Business Analytics & Data Mining
 Data Mining is an important sub task of Business
Analytics
 Both Predictive analysis and clustering tasks
utilize information retrieved from data mining
 Data mining helps handle some of the specific
problems faced when conducting Business
Analytics
 Dealing with and sorting through large data sets

Data Mining : An Overview
 What is Data Mining ?
 History
 Applications of Data Mining
 Detecting data discrepancies or outliers
 Relationship identification
 Data-Function mapping for modeling/prediction
 Categorizing and Summarizing Data
 Standards
 Challenges

Data Mining : What is it?
 Applying statistical analysis techniques to data
 the goal often being to determine unnoticed patterns or to
collect categorized information
 turns collected data into understandable structures
 Data Mining is often used as a buzz word to describe
processing large amounts of data
 In essence, its correct use relates to discovery of new
things through observation
 Synonymous with knowledge discovery

Data Mining : History
 Though HNC trademarked the term in 1990, hands on
pattern extraction is centuries old
 As long as statistic analysis has existed
 Discoveries in computer science have increasingly
shifted the field from hands on to machine dependent,
this allows for :
 The use of data indexing and DB systems to handle data
efficiently
 The application of statistical algorithms on a large scale,
possibly in a distributed manner, with less error

Data Mining : Use : Application
 Data Mining is often broken into several different
categories of tasks
 Detecting data discrepancies or outliers
 Relationship identification
 Data-Function mapping for modeling/prediction
 Categorizing and Summarizing Data

Data Mining : Finding outliers
 The process of analyzing large, mostly
homogeneous, sets of data and determining
which sets or points
 “go with the flow” and conform with patterns the rest
of the data seem to follow
 do not follow expected results when viewed against
the entire set of data
 An outlier can be a point or set of points, but can
also be defined through other means
 A period of time could yield unexpected results
 Ex. Network Intrusion

Data Mining : Techniques in finding outliers
 Rule Based – deciding a set of rules that
determine an outlier (or what isn’t one)
 Can be fuzzy or hard rules
 Cluster Analysis – As mentioned earlier
 Distance or Standard Deviation – Determining an
average over a data set and marking points that
aren’t within a Deviation or Distance

Applications of Outlier Detection
 Network Intrusion Detection
 Unusual bursts of network activity
 Identity Theft Detection
 Unusual spending or customer activity
 Detecting Software bugs
 Software does not deliver expected outputs
 Sensor event detection
 Monitoring patient health fluctuations in a medical setting
 Preprocessing
 Removing data skews based on extenuating
circumstances

Relationship Discovery: Basics
 Understanding how data is related is a key factor
in trend and knowledge discovery
 This is the definition of data mining
 Ex: Which products are often bought before a major
forecasted storm
 {hamburger buns} => {???}
 With small sets of data, or with correlations that
aren’t subtle (as the one above), identifying
relationships is not as difficult
 With large data sets or subtle relations a
combination of rule generation and data analysis
can be used to expedite the process

Relationship Discovery: How its done
 Since the number of relationships between points
of data could be boundless, two important
concepts are often introduced in relationship
discovery:
 The amount of data within which a relationship
might exist, called the support of a rule.
 The probability that data in the support will verify a
selected rule, called the confidence of a rule.

Relationship Discovery: How its done
 Generally we apply minimum bounds to both the support of
a rule and its confidence to determine relationships
 First : determine possible relationships
 Set a minimum support
 Orders with hamburgers, Orders with hamburger buns
 Other, user specific rules can be used here
 Second : take the remaining sets, look for patterns in the
items sets such that occurrence rate is above the minimum
confidence
 How many people bought hamburgers and buns together
 Ex: we find that if the customer is a male, and they buy
diapers, they will also buy beer
 {male, diapers} => {beer}

Matching data to functions
 Often, it is desirable to match data sets and the
factors that determine them to functions
 Allows for the possibility of predicting future results
 Involves learning how dependent and
independent variables in our data interact
 Dependent : the result, or where a point exists
 Independent : an cause or circumstance that
determines the dependent variable
 If we know how dependent and independent
variables interact, we can create a function and
run simulations to see results

Uses of Function-Data Mapping
 Weather Forecasting
 Determining what conditions lead to what kinds of
weather
 Stock market analysis
 When to buy and when to sell
 Crime Prevention
 What conditions cause or prevent crime

Categorizing
 Categorizing – Often we want to separate data
based off of a set of predefined attributes
 Very helpful in pattern recognition
 Ex: a persons political preference
 The process :
 we synthetically generate or measure a set of
observations (data points) with known categories
 we extract properties from said observations which
we believe contribute to the category
 These are called explanatory variables
 Finally we examine new data for these properties

Summarizing
 Summarizing – we almost never want to look at all of
the data individually
 Having too much data can actually hider the decision
making process
 Known as information overload
 Summarizing takes the results from data mining and
transforms it into formats that can be easily read
without omitting important information
 Summarizing might :
 Extract and display only important data
 correlate and abstract data to display trends
 Formats Include : Reports, Graphs, Dashboards, etc.

Standards : CRISP-DM
 Cross Industry Standard Process for Data Mining
 describes common practice for conducting data mining in an
enterprise setting
 KD nuggets – a community resource in DM and analytics
took polls and found CRISP-DM was the top methodology
in 02’, 04’, & 07’
 Six step methodology
 Business Understanding
 Data Understanding
 Data Preparation
 Modeling
 Evaluation
 Deployment

CRISP-DM : Explained
 Business Understanding
 Determining the business purpose
 Define success conditions – how do we know we succeeded
 Ex : improved prediction accuracy
 Map purpose/success conditions to data mining results
 Ex: fraud prevention => detect deviations
 Data Understanding
 Collecting and exploring data – defining its attributes
 Data quality verification

 Data Preparation
 Data Cleaning
 Normalization – fitting data within ranges
 Outlier removal – removing cases that could skew the model
 Handle missing attributes – the data was not obtained
 Formatting – changing data so that it fits with our tools
 Modeling – fitting the data to a model following the
methods previously described and then interpreting that
model
 Assess the accuracy of the collected data
 General purpose divided into prediction or description

 Evaluation – look at results and measure them with respect
to the success cases defined earlier
 Determine if one has succeeded
 Determine next steps, how do we apply the results
 Deployment – The execution of a strategy for using the
results of our data mining
 Includes preparing ways to monitor and maintain the
application of data mining results in the day to day
 Includes some sort of final summary

SEMMA
 Sample, Explore, Modify, Model and Assess
 Proposed by SAS Institute : A producer of BI and BA
software suites.
 Though this model is often considered general SAS
prefers to apply it directly to their products
 Focuses mainly on data mining and not on applying results
to business (unlike CRISP-DM)
Sampl
e
selecting the data set
Explor
e
Understand data through discovering relationships, both expected and
otherwise
Modify Transform and clean the data in order to prepare it for the modeling
process
Model Apply models to the data in order to discover trends and make predictions
Assess Evaluate the results of the modeling process to determine the reliability of
the mined data

Challenges in data mining
 Not enough or too much data
 Oftentimes it is difficult to access sufficient quantities of data
for small enterprises
 If the enterprise is large however, sometimes there is too
much and deciding what to keep is difficult
 Acquiring clean data
 Multiple formats or no format at all
 Privacy and ethical concerns
 Data aggregation : data compiled from multiple sources can
lead to revelations that violate privacy concerns
 Ex: anonymous data is collected and aggregated, leading to
identification

Business analytics and data mining

In this document