Predictive Analytics

Dr. Ramchandra Mangrulkar

 Analytics

Finding key customer characteristics and using those characteristics to

generate a score that could be used to determine which customers to mail.

2/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Credit Scoring in Financial Services

Credit Scoring
Predictive analytics is widely used in the financial sector to assess the
creditworthiness of individuals.

3/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Credit Scoring in Financial Services

Credit Scoring
Predictive analytics is widely used in the financial sector to assess the
creditworthiness of individuals.

Credit Score
Models analyze an individual’s credit history, payment behavior, and other
factors to predict the likelihood of timely repayment.

3/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Classification

4/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Disease Prediction in Healthcare

Healthcare Application
Predictive analytics is used in healthcare to predict the likelihood of diseases
or medical conditions in individuals.

5/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Disease Prediction in Healthcare

Healthcare Application
Predictive analytics is used in healthcare to predict the likelihood of diseases
or medical conditions in individuals.

Example
Models can analyze patient data to predict the risk of developing conditions
such as diabetes, heart disease, or cancer.

5/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Disease Prediction in Healthcare

6/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Inventory Management in Retail

Retail Application
Predictive analytics is used in retail for inventory management to forecast
demand and optimize stock levels.

7/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Inventory Management in Retail

Retail Application
Predictive analytics is used in retail for inventory management to forecast
demand and optimize stock levels.

Example
By analyzing historical sales data, retailers can make informed decisions about
stocking and restocking items, reducing stockouts or overstock situations.

7/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Inventory Management in Retail

8/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Employee Attrition Prediction in Human Resources

HR Application
Predictive analytics is applied in HR to predict employee turnover and identify
individuals at a higher risk of leaving.

9/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Employee Attrition Prediction in Human Resources

HR Application
Predictive analytics is applied in HR to predict employee turnover and identify
individuals at a higher risk of leaving.

Example
Analysis of factors such as employee performance, job satisfaction, and
historical turnover data helps organizations take proactive measures to retain
valuable employees.

9/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Traffic Flow Prediction in Transportation

Transportation Application
Predictive analytics is used in transportation to predict traffic patterns and
optimize routes.

10/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Traffic Flow Prediction in Transportation

Transportation Application
Predictive analytics is used in transportation to predict traffic patterns and
optimize routes.

Example
By analyzing historical traffic data, weather conditions, and other variables,
transportation systems can better plan for and manage traffic flow.

10/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Analytics Definition and Overview

Alert : Understanding Analytics

Analytics is the dynamic process of using computational methods to
reveal significant patterns in data.
The primary goal is to extract valuable insights, influencing informed
decision-making.
Analytics involves scrutinizing historical data, representing a measure of
past information.
The term ”analytics” gained prominence in 2005, with the advent of
Google Analytics.
Core concepts have roots in various terms such as cybernetics, data
analysis, neural networks, pattern recognition, statistics, knowledge
discovery, data mining, and data science.
The recent surge in analytics is driven by the practical need to enhance
estimates, forecasts, decisions, and organizational efficiency.

11/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Predictive Analytics Overview

Alert : Understanding Predictive Analytics

Predictive analytics is the process of discovering meaningful patterns in data.
Draws from disciplines like pattern recognition, statistics, machine learning,
artificial intelligence, and data mining.
What differentiates predictive analytics :
1. Data-driven : Algorithms derive model characteristics from data, not
analyst assumptions.
2. Automation : Algorithms automate pattern discovery, model building, and
input transformations.
Automation tasks include identifying variables, defining parameters, and
adjusting model complexity.
Transformations : Software automates the process of transforming input
variables for effective use in models.
Predictive analytics doesn’t replace human analysis but automates and
enhances complex tasks.
Challenges for analysts without automation : examining numerous histograms
and scatter plots for pattern identification.
Powerful aspect : Identifying the most important contributors to patterns in the
12/30 11.10.2023 data.
Dr. Ramchandra Mangrulkar DJSCE, Mumbai
 Supervised vs. Unsupervised Learning

Alert : Understanding Learning Paradigms

Predictive modeling algorithms are categorized into :
1. Supervised learning methods
2. Unsupervised learning methods
In **supervised learning** :
• Supervisor is the target variable, representing values predicted from other
data columns.
• Primary algorithms : Classification (categorical target) and Regression
(continuous target).
• Examples of target variables and applications.
In **unsupervised learning** :
• Also known as descriptive modeling.
• No target variable ; inputs are grouped or clustered based on proximity.
• Common in applications like customer analytics (segmentation).
Supervised learning assumes known inputs, while target variables may be
unobserved or unknown in some cases.
Common scenarios include predicting future events or behaviors (e.g.,
response models, cross-sell, and up-sell models).
13/30 11.10.2023 Predictive
Dr. Ramchandraanalytics
Mangrulkar not exclusively about predicting
DJSCE, Mumbai the future ; it can
 Supervised Learning Examples

Alert : Supervised Learning Scenarios

1. Classification :
• Identifying whether an email is spam or not spam.
• Predicting whether a student will pass or fail based on study hours.
• Diagnosing a medical condition (e.g., disease present or not).
2. Regression :
• Estimating the price of a house based on features like size and location.
• Predicting the stock price of a company based on historical data.
• Forecasting the sales of a product based on marketing spend.

14/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Classification

15/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Regression

16/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Unsupervised Learning Examples

Alert : Unsupervised Learning Scenarios

1. Clustering :
• Grouping customers based on purchasing behavior.
• Segmenting news articles into topics without predefined categories.
• Identifying patterns in stock market data for portfolio management.
2. Association :
• Recommending products based on shopping basket analysis.
• Discovering relationships between different variables in a dataset.
• Analyzing web click patterns to optimize content placement.

17/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Supervised Learning Business Models

Alert : Solving Problems with Supervised Learning

1. **E-commerce Recommendation System :**
• Utilize supervised learning to recommend products based on user
preferences and purchase history.
• Predict customer preferences and increase sales through personalized
recommendations.
2. **Credit Scoring for Loans :**
• Use supervised learning to assess creditworthiness and predict the
likelihood of loan default.
• Enable financial institutions to make informed lending decisions, reducing
risks.
3. **Medical Diagnosis :**
• Employ supervised learning models to predict diseases based on patient
symptoms and medical history.
• Assist healthcare professionals in early and accurate diagnosis, improving
patient outcomes.

18/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Unsupervised Learning Business Models

Alert : Solving Problems with Unsupervised Learning

1. **Customer Segmentation :**
• Apply unsupervised learning to group customers based on behavior,
preferences, and demographics.
• Tailor marketing strategies for different customer segments, optimizing
engagement.
2. **Anomaly Detection in Financial Transactions :**
• Utilize unsupervised learning to identify unusual patterns in financial
transactions.
• Enhance fraud detection systems by flagging abnormal activities for further
investigation.
3. **Content Recommendation for Media Platforms :**
• Use unsupervised learning to cluster content and recommend related media
to users.
• Improve user engagement by offering personalized content suggestions.

19/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Parametric vs. Non-Parametric Models

Parametric Models (E.g. Linear Regression)

Assume known distributions in the data.
Many parametric algorithms assume normal distributions and linear
relationships.
Advantage : If distributions are known, specific properties related to
errors, convergence, and certainty of learned coefficients can be proven.
Disadvantage : Requires data transformation, and assumptions may not
always hold.

Non-Parametric Models (E.g. Non-Parametric Models

Do not assume distributions in the data (distribution-free models).
More flexible as they don’t have underlying assumptions about data distribution.
Advantage : Saves time in preparing data, no need for distribution assumptions.
Disadvantage : Less known about the data a priori, often iterative without
guaranteed optimal solutions.

20/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Business Intelligence and Predictive Analytics

Overview
Business intelligence (BI) is a vast field that encompasses various analyses,
often resulting in reports or dashboards containing Key Performance Indicators
(KPIs).

Typical BI Output
Typical BI output includes reports for analysts and decision-makers,
addressing questions related to fraud detection and customer analytics.

Example Questions for Fraud Detection

How many cases were investigated last month ?
What was the success rate in collecting debts ?
How much revenue was recovered through collections ?
What was the ROI for various collection avenues (letters, calls, agents) ?
What was the close rate of cases in the past month, quarter, and year ?
For closed-out debts, how many days did it take on average to close them
21/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai
 Predictive Analytics vs. Business Intelligence - Fraud

Detection
Predictive Analytics Questions Business Intelligence Questions

What is the likelihood that How many cases were

the transaction is fraudulent ? investigated last month ?
What is the likelihood the What was the success rate in
invoice is fraudulent or collecting debts ?
warrants further How much revenue was
investigation ? recovered through collections ?
Which characteristics of the What was the ROI for various
transaction are most related collection avenues (letters,
to or most predictive of fraud calls, agents) ?
(single characteristics and
What was the close rate of
interactions) ?
cases in the past month,
What is the expected amount quarter, and year ?
of fraud ?
For closed-out debts, how
What is the likelihood that a many days did it take on
22/30 11.10.2023
tax return is non-compliant ?
Dr. Ramchandra Mangrulkar average
DJSCE, Mumbaito close them out ?
 Predictive Analytics vs. Business Intelligence - Customer

Analytics
Predictive Analytics Questions Business Intelligence Questions

What is the likelihood an What were the e-mail open,

e-mail will be opened ? click-through, and response
What is the likelihood a rates ?
customer will click-through a Which regions/states/ZIPs
link in an e-mail ? had the highest response
Which product is a customer rates ?
most likely to purchase if Which products had the
given the choice ? highest/lowest click-through
How many e-mails should the rates ?
customer receive to maximize How many repeat purchasers
the likelihood of a purchase ? were there last month ?
What is the best product to How many new subscriptions
up-sell to the customer after to the loyalty program were
they purchase a product ? there ?
23/30 11.10.2023 What is the
Dr. Ramchandra visit volume
Mangrulkar What is the average spend of
DJSCE, Mumbai
 Key Differences

Differences in Questions
Notice the differences in the kinds of questions predictive analytics asks
compared to business intelligence. The word “likelihood” appears often,
meaning we are computing a probability that the pattern exists for a unit of
analysis.

Interpretation
If the likelihood an individual customer will purchase a product is one percent,
this means that for every 100 customers with the same pattern of measured
attributes for this customer, one customer purchased the product in the
historic data used to compute the likelihood.

User-Driven vs. Data-Driven

The difference between the business intelligence and predictive analytics measures is that the business intelligence variables identified in
the questions were user-driven. In the predictive analytics approach, the predictive modeling algorithms considered many patterns,
sometimes all possible patterns, and determined which ones were most predictive of the measure of interest (likelihood). The discovery
of the patterns is data-driven.

24/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Predictive Analytics - Future Prediction

Predictive Analytics as Future Prediction

Predictive analytics is often associated with forecasting future behavior.
Questions like ”What is the likelihood a customer will purchase...” imply a
focus on predicting future actions.

25/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Timeline for Data Usage

Timeline for Data Usage

Figure 1-2 illustrates a timeline related to data used for building predictive
models or business intelligence reports. Both utilize historic data, with
predictive models specifically focusing on events in the future.

Common Data Source

It’s crucial to note that both business intelligence and predictive analytics
analyses are constructed from the same historic data. The assumption is that
future behavior will be consistent with past behavior.

26/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Modeling Data Timeframe

Modeling Data Timeframe

When building predictive models, the data is set up with all the modeling data
in the past, but the target variable (future behavior) is considered future to
the date the attributes are collected.

27/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Consistency in Future Behavior

Consistency in Future Behavior

The assumption is that patterns identified by predictive models in the past will
continue to hold in the future, establishing consistency between past and
future behavior.

28/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Statistics vs. Predictive Analytics

STATISTICS PREDICTIVE ANALYTICS

Models based on theory : There is Models often based on non-
an optimum. parametric algorithms ; no guaran-
teed optimum.
Models typically linear. Models typically nonlinear.
Data typically smaller ; algorithms Scales to big data ; algorithms not
often geared toward accuracy with as efficient or stable for small data.
small data.
The model is king. Data is king.

29/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai

 Summary : Predictive Analytics vs. Statistics

**Overlap with Statistics :** - Predictive analytics is considered an

extension of statistics. - Predictive modelers often use statistical
algorithms and tests in their techniques.
**Algorithm Usage :** - Diagnostics application in predictive analytics
may differ from traditional statistical practices.
**Importance of Statistics in Predictive Analytics :** - Predictive
analytics draws heavily from statistical principles. - Acknowledging the
quote from statistician Brad Efron : ”Those who ignore Statistics are
condemned to reinvent it.”
**Differences Between Fields :** - Significant differences exist in typical
approaches between statistics and predictive analytics.
**Role of Theory :** - Statistics is driven by theory, while predictive
analytics often adopts algorithms from machine learning and artificial
intelligence.
**Fundamental Difference :** - For statistics, the model is king. - For
predictive analytics, data is king.
30/30 11.10.2023 Dr. Ramchandra Mangrulkar DJSCE, Mumbai