Module 4:

Predictive Analytics I - Data mining process, methods, and Algorithms

Lecture Notes – 5.2 Data Mining Concepts and Applications

1. Introduction

• Data Mining: Process of discovering hidden patterns, trends, and useful

knowledge from large datasets.

• Historical Context:

o Dr. Arno Penzias (1999): “Data mining will become the key corporate
application of the future.”

o Thomas Davenport (2006): Analytics = strategic weapon for firms

(Amazon, Capital One, Marriott).

• Why Important?

o Helps companies understand customers, vendors, supply chains → better

decision-making, profitability, and competitive advantage.

2. Growth of Data Mining

• Reasons for Emergence:

o Global competition → need for better customer insights.

o Untapped value hidden in large datasets.

o Consolidation of databases → creation of data warehouses.

o Explosion of data storage & processing technologies.

o Decrease in hardware/software costs.

o Rise of the Internet → massive data generation (social, genomic,

astronomy, medical, etc.).

3. Applications of Data Mining

• Finance: Fraud detection (credit cards, insurance).

• Retail: Market-basket analysis (buying beer & diapers together).

 • Healthcare: Medical diagnosis, drug discovery.

• E-commerce: Customer targeting, recommendation systems.

• Other domains: Astronomy, nuclear physics, pharmaceuticals, genomics.

4. Definition of Data Mining

• Simple definition: Discovering knowledge from large datasets.

• Formal definition (Fayyad et al., 1996):

“Nontrivial process of identifying valid, novel, potentially useful, and ultimately
understandable patterns in structured data.”

• Key Properties:

o Process: Iterative, multi-step.

o Nontrivial: Involves exploration/inference (not simple calculations).

o Valid: Patterns hold true on new data.

o Novel: Previously unknown.

o Useful: Must lead to benefits.

o Understandable: Must make business sense.

• Alternative Terms: Knowledge discovery, pattern analysis, information

harvesting, data archaeology.

5. Characteristics of Data Mining

• Works with large, complex datasets (often in data warehouses).

• Operates in client/server or web-based IS environments.

• Uses advanced visualization tools for unstructured & structured data.

• Empowers end users with minimal programming skills.

• Often requires parallel processing due to massive data volumes.

• Integrates with spreadsheets and decision support tools.

6. Data Mining Process & Models

• Goal: Extract patterns & build models.

• Types of Patterns:

1. Associations – co-occurrence (beer & diapers).

2. Predictions – future trends/events (weather forecast).

3. Clusters – natural groupings (customer segmentation).

4. Sequential patterns – time-ordered events (checking → savings →

investment account).

• Learning Approaches:

o Supervised learning: Training data has input + output (classification,

regression).

o Unsupervised learning: Only input data, no class labels (clustering,

associations).
7. Data Mining Tasks

A. Prediction

• Definition: Foretelling future values using past data.

• Types:

o Classification (categorical output, e.g., “spam” vs. “not spam”).

o Regression (continuous output, e.g., predicting stock prices).

• Techniques:

o Neural Networks, Decision Trees, SVM, k-NN, Logistic Regression.

B. Clustering

• Definition: Grouping objects without predefined labels.

• Goal: Maximize intra-group similarity, minimize inter-group similarity.

• Techniques: k-means, DBSCAN, Self-Organizing Maps (SOM), EM algorithm.

• Applications: Market segmentation, fraud detection.

C. Associations

• Definition: Discovering relationships among items.

• Applications: Market-basket analysis, link analysis, sequence mining.

• Algorithms: Apriori, FP-Growth, Eclat, OneR, ZeroR.

8. Visualization & Time-Series Forecasting

• Visualization: Helps interpret mined patterns; visual analytics combines

visualization + analytics.

• Time-series forecasting: Uses past sequential data (e.g., stock prices,

temperature) to predict future values.

• Techniques: ARIMA, exponential smoothing, autoregressive models.

9. Data Mining vs. Statistics

Aspect Statistics Data Mining

Approach Hypothesis-driven Discovery-driven

Data Type Sample data (primary) Full datasets (secondary)

Scale Small to medium (100–1000s) Very large (millions–billions)

Goal Test hypothesis Discover hidden patterns

Method Mathematical/statistical models AI, ML, pattern recognition + statistics

10. Strategic Benefits of Data Mining

• Enhanced decision-making.

• Improved customer relationship management (CRM).

• Fraud detection & risk management.

• Increased profitability via targeted marketing.

• Competitive advantage in saturated markets.

Summary:
Data mining = interdisciplinary process combining statistics, AI, machine learning, and
databases to discover hidden knowledge from massive datasets. It supports prediction,
classification, clustering, and association, making it invaluable in domains like finance,
retail, healthcare, and e-commerce.

5.3 Data Mining Applications – Lecture Notes

Introduction

• Data mining is widely applied to address complex business problems and

opportunities.

• Goal → create sustainable competitive advantage by solving pressing issues or

exploring new opportunities.

Major Application Areas

1. Customer Relationship Management (CRM)

• Builds one-on-one relationships with customers.

• Applications:

o Customer profiling (identify likely buyers).

o Churn analysis (understand attrition).

o Discover product/service associations.

o Identify most profitable customers & strengthen loyalty.

2. Banking

• Automates loan approval & predicts defaults.

• Detects fraudulent credit card & online transactions.

• Cross-selling: products/services to right customers.

• Forecasts cash flow at ATMs & branches.

3. Retailing & Logistics

• Forecasts sales volumes to set inventory levels.

• Market-basket analysis → store layout & promotions.

• Predicts seasonal demand for products.

• Tracks product movement (RFID, sensors) → manage perishables.

4. Manufacturing & Production

• Predicts machine failures (condition-based maintenance).

• Detects anomalies to optimize production.

• Improves product quality through pattern discovery.

5. Brokerage & Securities Trading

• Predicts bond price changes & stock fluctuations.

• Analyzes effects of events on market.

• Detects & prevents fraudulent trading activities.

6. Insurance

• Forecasts claim amounts (health, property).

• Determines optimal rate plans.

• Predicts likely buyers of new policies.

• Prevents fraudulent or incorrect claim payments.

7. Computer Hardware & Software

• Predicts disk drive failures.

• Filters spam & unwanted web content.

• Detects network intrusions/security breaches.

• Identifies insecure software.

8. Government & Defense

• Forecasts cost of military movements.

• Predicts adversary moves & strategy planning.

• Resource planning & budgeting.

• Knowledge sharing: lessons from operations.

9. Travel Industry (Airlines, Hotels, Rentals)

• Yield management → optimize pricing dynamically.

• Forecasts demand at different locations.

• Identifies profitable customers → personalized services.

• Reduces employee attrition through root-cause analysis.

10. Healthcare

• Identifies uninsured populations & causes.

• Finds cost–benefit patterns among treatments.

• Forecasts demand at hospitals/clinics.

• Studies causes of customer & employee attrition.

11. Medicine

• Improves cancer survivability with novel patterns.

• Predicts organ transplant success → better donor matching.

 • Genomics → gene function identification.

• Maps illness-symptom-treatment relationships.

12. Entertainment Industry

• Analyzes viewer data → show scheduling & ads placement.

• Predicts movie success before production.

• Forecasts demand for events → scheduling & resource allocation.

• Dynamic pricing policies to maximize revenue.

13. Homeland Security & Law Enforcement

• Identifies terrorist behavior patterns.

• Detects crime patterns (time, location, behavior).

• Predicts & prevents biological/chemical attacks.

• Protects critical information infrastructures.

14. Sports

• NBA, MLB, NCAA → predictive analytics for performance.

• Optimizes resource allocation for winning.

• Predicts tournament outcomes (March Madness, Bowl Games).

• Enhances coaching and player recruitment strategies.

Summary

• Data mining applications span business, healthcare, defense, entertainment,

and sports.

• Key benefits: prediction, fraud detection, personalization, optimization,

competitive advantage.

• Data mining has become an indispensable decision-making tool across

industries.
Lecture Notes – 5.4 Data Mining Process

Introduction

• Data mining projects need a systematic process for consistency and success.

• Several standardized methodologies exist:

o CRISP-DM (Cross-Industry Standard Process for Data Mining) – most

popular.

o SEMMA (Sample, Explore, Modify, Model, Assess) – developed by SAS

Institute.

o KDD (Knowledge Discovery in Databases) – broader framework.

CRISP-DM Process (Cross-Industry Standard Process for Data Mining)

• Developed in mid-1990s by European consortium.

• Six-step methodology (iterative, not strictly linear).

• Focus: From business understanding → deployment of data mining solution.

Step 1: Business Understanding

• Define objective of the study.

• Example questions:

o Why are we losing customers to competitors?

o What is the lifetime value of each customer?

• Create a project plan: roles, responsibilities, timelines, and budget.

Step 2: Data Understanding

• Identify relevant data sources for the business task.

• Activities:

o Collect, describe, and explore data.

o Statistical & visualization methods: histograms, scatterplots, boxplots,

correlations.

• Data types:

o Quantitative (numeric: discrete/continuous).

o Qualitative (categorical: nominal/ordinal).

• Sources: demographics, sociographics, transactions, social media, machine-

generated data.

Step 3: Data Preparation (Preprocessing)

• Consumes ~80% of total project time.

• Goal: Prepare clean and consistent data.

• Issues addressed:

o Incomplete data (missing values, missing attributes).

o Noisy data (errors, outliers).

o Inconsistent data (duplicate/conflicting codes).

• Techniques: data cleaning, transformation, normalization, integration.

Step 4: Model Building

 • Apply different modeling techniques (classification, regression, clustering,
association).

• Compare models to identify the best performer.

• Algorithms:

o Decision Trees → Classification.

o K-means → Clustering.

o Apriori → Association rules.

• Often requires going back to data preparation (formatting issues).

Step 5: Testing & Evaluation

• Evaluate accuracy and generalizability of models.

• Verify alignment with business objectives.

• Methods:

o Cross-validation, test sets, performance metrics.

o Visualization and tabulation: pivot tables, pie charts, histograms,

scatterplots.

• Requires collaboration between:

o Data analysts (technical results).

o Business analysts (practical interpretation).

o Managers (decision-making).

Step 6: Deployment

• Deliver final results to end users.

• Could be:

o A simple report.

o Or a repeatable enterprise-wide data mining solution.

• Includes:

o Training users.
 o Monitoring and maintaining models (since business data changes over
time).

• Maintenance strategy essential to avoid outdated/incorrect models.

Other Methodologies

1. SEMMA (by SAS Institute)

• Steps:

1. Sample – generate representative sample of data.

 2. Explore – visualize and understand data distributions.

3. Modify – select & transform variables.

4. Model – build predictive models.

5. Assess – evaluate accuracy and usefulness.

• Iterative experimentation cycle.

• Difference from CRISP-DM:

o CRISP-DM includes business understanding.

o SEMMA assumes business goals & data sources are already defined.

2. KDD (Knowledge Discovery in Databases)

• Defined by Fayyad et al. (1996).

• Broader process → data mining is one step.

• Steps:

1. Data Selection.

2. Data Cleaning (preprocessing).

3. Data Transformation.

4. Data Mining (pattern discovery using algorithms).

5. Interpretation & Evaluation (convert patterns into actionable knowledge).

• Criticism: Treats data mining as a single step, not the whole process.

Adoption Ranking of Data Mining Methodologies

• KDnuggets survey (2007):

 o CRISP-DM → most widely used.

o SEMMA and KDD → less common, but still influential.

Key Takeaways

• Data mining projects must follow a standard process.

• CRISP-DM is the most comprehensive & widely adopted.

• SEMMA focuses more on technical data preparation and modeling.

• KDD is a broader concept, viewing data mining as part of overall knowledge

discovery.

• Successful projects require collaboration between technical experts &

business stakeholders.
Lecture Notes – 5.5 Data Mining Methods

Overview

Data mining methods are used to discover patterns, relationships, and

knowledge from large datasets.
The most common techniques include:

• Classification

• Regression

• Clustering

• Association Rule Mining

Most data mining tools use multiple algorithms for these methods.

1. Classification

Definition

Classification is a supervised learning technique that learns patterns from

labeled data to assign new data into predefined classes.

Example:
Predicting weather type (“Sunny,” “Rainy,” “Cloudy”) based on humidity,
temperature, etc.

Applications

• Credit approval (Good/Bad risk)

• Target marketing (Likely/Unlikely customer)

• Fraud detection (Yes/No)

• Telecom churn prediction (Switch/Stay)

Classification Process

1. Model Development (Training)

o Learn from historical labeled data.

2. Model Testing and Deployment

o Evaluate using unseen data (holdout/test data).

 o Deploy for future prediction.

Model Evaluation Metrics

Metric Formula Description

(TP + TN) / (TP + Percentage of correct

Accuracy
TN + FP + FN) predictions

True Positive Rate Correctly identified

TP / (TP + FN)
(Sensitivity) positives

True Negative Rate Correctly identified

TN / (TN + FP)
(Specificity) negatives

Fraction of correct
Precision TP / (TP + FP)
positive predictions

Fraction of positives
Recall TP / (TP + FN)
correctly identified

(TP = True Positive, TN = True Negative, FP = False Positive, FN = False Negative)

Confusion Matrix Example:

Actual / Predicted Positive Negative

Positive TP FN

Negative FP TN

Model Accuracy Estimation Methods

1. Simple Split (Holdout)

o Divide data: 2/3 training, 1/3 testing

o Problem: random partition may cause bias

2. K-Fold Cross Validation

o Split data into k parts (e.g., k = 10)

o Train on (k–1) folds, test on the remaining fold

o Average accuracy across all folds

o More reliable than simple split

3. Other Techniques

o Leave-One-Out → Each record tested once.

o Bootstrapping → Sampling with replacement.

o Jackknifing → Similar to leave-one-out.

o ROC Curve (AUC) → Graphical measure of model performance.

 ▪ AUC = 1 → Perfect model

▪ AUC = 0.5 → Random chance

Classification Algorithms

Category Description / Example

Most popular; uses tree-like structures to

Decision Trees
classify data

Statistical Methods Logistic Regression, Discriminant Analysis

Neural Networks Learn nonlinear relationships

Case-Based
Compares new case with similar past cases
Reasoning

Bayesian Classifiers Probabilistic model based on prior occurrences

Search-based classification inspired by

Genetic Algorithms
evolution

Rough Sets Handle uncertainty in classification

 2. Ensemble Models

Concept

Combining multiple models improves accuracy, robustness, and reduces bias.

Types

1. Homogeneous Ensembles

o Combine same model types (e.g., multiple decision trees)

o Examples:

▪ Random Forest (Bagging)

▪ AdaBoost (Boosting)

2. Heterogeneous Ensembles

o Combine different model types (e.g., Decision Tree + SVM + Neural

Network)

o Also called Information Fusion Models

Combination Methods:

• Voting (equal weight)

• Weighted Voting (based on accuracy)

 3. Decision Trees

Key Terms

• Attributes: Input variables (e.g., income, credit rating)

• Nodes & Branches: Represent tests on attributes

• Leaf Nodes: Represent class labels (final outcomes)

Process

1. Create a root node with training data

2. Select the best splitting attribute

3. Split data into branches

4. Repeat recursively until stopping condition (pure node) is met

Splitting Measures

1. Gini Index (CART Algorithm)

o Measures impurity of a node

o Smaller Gini → Better split

𝑛

𝐺𝑖𝑛𝑖(𝑆) = 1 − ∑ 𝑝𝑗2
𝑗=1

2. Information Gain (ID3/C4.5 Algorithm)

o Uses entropy to measure uncertainty

o Higher Gain → Better attribute

𝐺𝑎𝑖𝑛(𝐴) = 𝐼(𝑝, 𝑛) − 𝐸(𝐴)

4. Cluster Analysis

Definition

Unsupervised learning technique that groups similar items into clusters based
on similarity or distance measures.

Applications
• Market segmentation

• Fraud detection

• Image recognition

• Customer grouping

Types

1. Hierarchical Clustering

o Agglomerative (bottom-up)

o Divisive (top-down)

2. Partitioning Clustering

o e.g., k-means algorithm

K-Means Algorithm Steps

1. Choose number of clusters (k)

2. Randomly assign centroids

3. Assign each data point to nearest centroid

4. Recalculate centroids

5. Repeat until stable

Distance Measures

• Euclidean Distance

• Manhattan Distance

• Weighted distances (based on attributes)

 Choosing Number of Clusters

• Elbow Method (Variance explained curve)

• AIC (Akaike Information Criterion)

• BIC (Bayesian Information Criterion)

• Rule of Thumb: √𝑛/2

5. Association Rule Mining (Affinity Analysis)

Definition

Finds relationships among items that frequently occur together.

Famous Example

Beer + Diapers pattern in retail transactions.

Applications

• Market basket analysis

• Cross-selling / promotions

• Web usage mining

• Fraud detection

• Healthcare diagnostics

Key Metrics

Metric Definition Purpose

Frequency of itemset
Support (S) Popularity of rule
occurrence

Confidence Probability that Y occurs when X

Reliability of rule
(C) occurs

Ratio of observed confidence to Strength of

Lift (L)
expected confidence association
 Example Rule:

{𝐿𝑎𝑝𝑡𝑜𝑝, 𝐴𝑛𝑡𝑖𝑣𝑖𝑟𝑢𝑠} ⇒ {𝐸𝑥𝑡𝑒𝑛𝑑𝑒𝑑𝑊𝑎𝑟𝑟𝑎𝑛𝑡𝑦}[𝑆𝑢𝑝𝑝𝑜𝑟𝑡 = 30%, 𝐶𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒

= 70%]

Applications by Domain

• Retail: Product placement, promotions

• Banking: Sequential service usage

• Insurance: Claim pattern detection

• Telecom: Service bundling

• Healthcare: Symptom-disease relations

APRIORI ALGORITHM

The Apriori algorithm is the most commonly used algorithm to discover

association rules.

Summary Table

Learning
Method Goal Example
Type

Predict
Classification Supervised Spam detection
class label
 Learning
Method Goal Example
Type

Predict
Regression Supervised numeric House price
value

Group
Customer
Clustering Unsupervised similar
segmentation
items

Discover
Market basket
Association Unsupervised item
analysis
relations

Key Takeaways

• Data mining transforms raw data into actionable insights.

• Choice of method depends on data type and business objective.

• Ensemble and hybrid models improve performance.

• Evaluation and validation are critical for model trustworthiness.

Prescriptive Analytics - Optimization and Simulation:

Lecture Notes: Model-Based Decision Making

1. Introduction to Model-Based Decision Making

Definition

• Model-based decision-making refers to using analytical or mathematical

models to guide decision processes.

• It is the core of Prescriptive Analytics, the third phase of the analytics maturity
model:

o Descriptive Analytics: What happened?

o Predictive Analytics: What is likely to happen?

o Prescriptive Analytics: What should we do?

Purpose

• To determine optimal actions or recommendations based on data-driven

insights.

• Moves from understanding and prediction to decision guidance or automation.

2. Role and Importance of Prescriptive Analytics

• Prescriptive analytics helps decision-makers:

o Optimize resources (e.g., cost, time, workforce).

o Maximize profits, efficiency, or customer satisfaction.

o Reduce uncertainty through data-supported decisions.

Example Scenarios

• Deciding which customers should receive offers or promotions to maximize ROI.

• Determining optimal warehouse locations to minimize logistics costs.

• Setting delivery routes or staffing schedules based on predicted demand.

• Choosing optimal bidding or pricing strategies in marketing and procurement.

3. Decision Analytics and Justification

• Model-based decisions provide:

o Consistency: Structured and repeatable decision process.

o Objectivity: Data-driven reasoning replaces intuition.

o Justification: Models provide evidence for decision recommendations.

• Thus, prescriptive analytics is often termed Decision Analytics.

4. Real-World Example: Canadian Football League (CFL) Case Study

Problem

• Scheduling 81 games for 9 teams over 5 months with multiple constraints:

o Team rest days.

o Maximizing sales revenue and TV ratings.

o Scheduling rivalry games on public holidays.

o Managing time-zone differences.

Solution

• Used Microsoft Excel Solver and Frontline Systems’ Premium Solver.

• Converted a nonlinear scheduling problem into a linear optimization model.

• Considered constraints and objectives simultaneously.

Results

• Improved player rest periods → Better performance.

• Higher ticket sales and TV ratings → Increased revenue.

• Optimized schedule satisfied major stakeholders.

Key Learning

• Even simple tools like Excel Solver can deliver prescriptive analytics value.

• Real-world applications can yield multi-million dollar benefits.

5. Real-World Example: Refinery Modeling (Downstream Advisors Case Study)

Problem

• Evaluating, simulating, and optimizing operations in oil refineries.

• Needed to understand flow rates, product routing, and process optimization.

Solution

• Developed Smart Ref, a customizable refinery model.

• Used Analytic Solver (Frontline Systems, Excel-based).

• Simulated multiple refinery operations with nonlinear and linear models.

Benefits

• Provided investors with insights into operational efficiency and ROI.

• Enabled optimization of refinery distribution and product flow.

• Supported faster decision-making and competitive advantage.

Lesson Learned

• Linear and nonlinear programming models can optimize complex industrial

processes.

• Solver tools can produce optimal solutions in seconds for large-scale

problems.

6. Steps in Prescriptive Analytics Modeling

a. Problem Identification & Environmental Analysis

• Understand the domain, stakeholders, and decision context.

• Analyze:

o Organizational culture.

o Decision-making hierarchy.

o External and internal environmental factors.

• Use Business Intelligence (BI) tools for environmental scanning and problem
detection.

b. Variable Identification

• Identify:
 o Decision variables: What we can control (e.g., price, quantity).

o Uncontrollable variables: External factors (e.g., demand, weather).

o Result variables: Outcomes or performance measures.

• Use Influence Diagrams and Cognitive Maps to visualize relationships.

c. Forecasting (Predictive Analytics Integration)

• Essential precursor to prescriptive analytics.

• Predict future demand, market trends, or consumer behavior.

• Enables “what-if” and sensitivity analyses for future decision outcomes.

7. Model Categories in Prescriptive Analytics

Category Process & Objective Representative Techniques

Optimization (few Find the best among few

Decision tables, trees, AHP
alternatives) options

Optimization (many Iterative search for best Linear & mathematical

alternatives) solution programming, network models

Analytic formula Solve using direct equations Inventory models

Experiment with models to

Simulation Monte Carlo, system dynamics
find acceptable solutions

Apply rule-based logic for Expert systems, heuristic

Heuristics
good solutions algorithms

Regression, forecasting, Markov

Predictive models Forecast future events
models

Explore hypothetical “what-if”

Other models Financial or queuing models
scenarios

8. Model Management and Knowledge-Based Modeling

• Model Management Systems (MMS):

o Manage model versions, data sources, and metadata.

 o Ensure model integrity and reusability (analogous to DBMS).

• Knowledge-Based Models:

o Combine quantitative (mathematical) and qualitative (expert)

reasoning.

o Use AI-based systems such as expert systems, rule engines, and

machine learning.

9. Current Trends in Modeling

a. Model Libraries & Open-Access Tools

• Libraries and solvers are now accessible online.

o Example: NEOS Server for Optimization (University of Wisconsin-

Madison)

o INFORMS Resources: For OR and management science models.

b. Cloud-Based Modeling and Simulation

• Cloud services allow large-scale model execution and storage.

• Examples: Google OR-Tools, AWS SageMaker, MATLAB Online.

c. Integration with Data Warehousing & Parallel Computing

• Large-scale prescriptive models depend on:

o Big data warehouses for data input.

o High-performance computing (HPC) for solution speed.

d. Transparency and Usability

• Emphasis on user-friendly, spreadsheet-like modeling tools (e.g., OLAP

systems).

• Multidimensional analysis allows decision-makers to interact with models

intuitively.

10. Summary and Key Takeaways

• Model-based decision-making = Prescriptive Analytics.

• It moves from predicting outcomes to recommending optimal actions.

• Tools like Excel Solver, Analytic Solver, and cloud-based platforms make
modeling accessible.

• Effective modeling requires:

o Clear problem definition.

o Correct variable identification.

o Integration of predictive insights.

• Applications span across:

o Sports scheduling, logistics, refinery operations, marketing, HR, and

finance.

• Prescriptive analytics is the frontier of intelligent decision support in modern

organizations.
Lecture Notes: Structure of Mathematical Models for Decision Support

1. Introduction

Mathematical models are at the core of decision support systems (DSS) and
prescriptive analytics.
They help decision-makers analyze, predict, and optimize outcomes in complex
environments such as business, finance, and engineering.

A mathematical model represents a real-world system using mathematical

relationships among various components—variables, parameters, and results.

2. Components of Decision Support Mathematical Models

All quantitative models generally consist of four main components:

1. Result (Outcome) Variables

2. Decision Variables

3. Uncontrollable Variables (Parameters)

4. Intermediate Result Variables

These are interconnected through mathematical relationships (equations or

inequalities).

2.1 Result (Outcome) Variables

• Represent the effectiveness or performance of a system.

• These are the outputs of the model.

• Also called dependent variables, since their values depend on other variables.

• Measure how well goals or objectives are achieved.

Examples:

• Total profit

• Rate of return on investment (ROI)

• Customer satisfaction

• Market share

• Total cost

Example Context:
In a manufacturing system, “total cost” or “profit” would be the result variable.

2.2 Decision Variables

• Represent the choices or actions that the decision-maker controls.

• Define alternative strategies or actions available for implementation.

• Changing their values alters the model outcomes.

Examples:

• Investment amount in bonds

• Advertising budget

• Production quantity

• Inventory levels

• Staffing levels

• Delivery schedule

Example Context:
In a transportation model, shipment routes and schedules are decision variables.

2.3 Uncontrollable Variables (Parameters)

• Represent factors that influence outcomes but cannot be controlled by the

decision maker.

• Often come from the external environment.

• Some may be constant (parameters), while others may vary (variables).

 • They often form constraints in the problem.

Examples:

• Inflation rate

• Prime interest rate

• Tax regulations

• Market demand

• Legal requirements

• Machine capacity

Example Context:
A manufacturer cannot control raw material prices or government tax policies.

2.4 Intermediate Result Variables

• Represent intermediate outcomes that occur between decision actions and

final results.

• Help in understanding cause–effect relationships within the model.

Examples:

• Spoilage (in production) → affects total profit.

• Employee satisfaction → affects productivity.

• Machine utilization → affects total cost.

Example Context:
In a factory model:
Decision variable = “Employee salary”
Intermediate variable = “Employee satisfaction”
Final result variable = “Productivity level”

3. Relationships Among Model Components

These components are connected through mathematical relationships, which can be:

• Equations – express equality (e.g., Profit = Revenue − Cost)

• Inequalities – represent constraints (e.g., Production ≤ Machine capacity)

Mathematical models can be:

 • Deterministic – all values are known with certainty.

• Stochastic – include random variables to represent uncertainty.

4. Structure of a Quantitative Model

A general model structure can be represented as:

Result Variables = f(Decision Variables, Uncontrollable Variables, Intermediate

Variables)

or graphically (as in Figure 8.1):

Uncontrollable Variables

Decision Variables → Intermediate Variables → Result Variables

These relationships define how inputs (decisions + environment) determine outputs

(results).

5. Examples of Mathematical Models

Example 1: Basic Profit Model

𝑃 =𝑅−𝐶

Where:

• 𝑃= Profit (Result variable)

• 𝑅= Revenue (Intermediate variable)

• 𝐶= Cost (Intermediate variable)

This is the simplest algebraic representation of a business model.

Example 2: Present Value Model (Finance)

𝐹
𝑃=
(1 + 𝑖)𝑛

Where:

• 𝑃= Present value
 • 𝐹= Future payment

• 𝑖= Interest rate

• 𝑛= Number of years

Example Calculation:
Determine the present value of ₹100,000 due in 5 years at a 10% interest rate:
100,000
𝑃= = 62,092
(1 + 0.1)5

This shows how time value of money is modeled mathematically for decision making.

6. Summary Table: Components of Models

Uncontrollable
Area Decision Variables Result Variables
Variables / Parameters

Financial Investment Inflation, tax rates,

ROI, profit
Investment alternatives, amount interest rate

Advertising budget, Competitor actions,

Marketing Market share, sales
media choice customer income

Production quantity, Total cost, quality Machine capacity,

Manufacturing
inventory levels level material prices

Audit schedule, Data processing Legal requirements,

Accounting
computer usage cost, error rate technology

Transport cost,
Delivery distance,
Transportation Shipment schedules customer
regulations
satisfaction

Staffing levels, Customer

Services Demand, competition
service design satisfaction

7. Key Takeaways

• Mathematical models support decision-making by quantifying relationships

among variables.
 • Decision variables represent controllable actions, while uncontrollable
variables capture environmental factors.

• Result variables indicate performance, and intermediate variables bridge

decisions and outcomes.

• Models can range from simple algebraic to complex optimization and

simulation models.

• The ultimate goal is to improve decisions through structured analysis.

Lecture Notes: Certainty, Uncertainty, and Risk

1. Introduction

Decision-making always involves evaluating alternatives and predicting future

outcomes.
However, the amount of information available about the future varies, leading to
different decision environments.

Decision situations are generally classified into three categories, depending on the
level of knowledge about future outcomes:

1. Decision-making under Certainty

2. Decision-making under Uncertainty

3. Decision-making under Risk

These categories represent a continuum of knowledge — from complete knowledge to

total ignorance.

Figure 8.2: The Zones of Decision-Making

 Decision
Knowledge Level Description
Environment

Complete
Certainty All outcomes are known
Knowledge

Probabilities of outcomes are known or

Partial Knowledge Risk
estimated

Very Limited Outcomes are unknown and probabilities

Uncertainty
Knowledge cannot be assigned

Graphically (conceptually):

Increasing Knowledge → Certainty → Risk → Uncertainty → Total Ignorance

2. Decision-Making under Certainty

Definition:

• The decision maker knows exactly what the outcome of each decision
alternative will be.

• There is complete information about all possible outcomes.

• The environment is deterministic (no randomness).

Characteristics:

• Single known outcome for each alternative.

• Simplifies analysis — models are easy to construct and solve.

• Applicable mostly to structured problems and short-term horizons (e.g., up to

1 year).

• Can produce optimal solutions because there is no ambiguity.

Example:

• Investment in U.S. Treasury Bills:

The return rate (e.g., 4%) is fixed if held to maturity.
Hence, the decision maker knows the exact outcome in advance.

Applications:

• Financial budgeting under fixed rates

• Deterministic inventory models

 • Short-term production scheduling

⚙ Note: Although complete certainty rarely exists in reality, assuming certainty often
helps simplify complex models for analysis.

3. Decision-Making under Uncertainty

Definition:

• The decision maker faces multiple possible outcomes for each alternative.

• The probability of these outcomes is unknown or cannot be reliably estimated.

• The environment is ambiguous or unpredictable.

Characteristics:

• Insufficient or incomplete information.

• The decision-maker cannot assign probabilities to outcomes.

• Requires subjective judgment or assumptions about behavior toward risk.

• Managers often try to reduce uncertainty by gathering more information.

Managerial Behavior:

• Many managers avoid uncertainty, trying to obtain more data to turn the
problem into a risk or certainty case.

• When information is unavailable, models must handle true uncertainty, often

through qualitative assessment or heuristic decision rules.

Example:

• Launching a new, innovative product with no historical data:

The company cannot predict market demand or consumer acceptance.

Approaches under Uncertainty:

• Optimistic (Maximax) approach: Choose the alternative with the best possible
outcome.

• Pessimistic (Maximin) approach: Choose the alternative with the least-worst

outcome.

• Equally likely (Laplace) criterion: Assume all outcomes are equally probable.
4. Decision-Making under Risk (Risk Analysis)

Definition:

• Multiple outcomes exist for each alternative.

• Probabilities of outcomes are known or can be estimated.

• The decision-maker can quantify uncertainty using probabilities — this is called

calculated risk.

• The environment is probabilistic or stochastic.

Characteristics:

• Probabilities may come from:

o Historical data

o Expert judgment

o Statistical estimation

• Allows quantitative analysis of risk and expected performance.

• Most business and financial decisions are made under assumed risk.

Example:

• Investing in stocks:
Possible outcomes include profit, small loss, or large loss — each with known or
estimated probabilities.

Risk Analysis Techniques:

• Expected Value (EV) Calculation:

𝐸𝑉 = ∑(𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑖 × 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖 )

Select the alternative with the highest expected value.

• Variance or Standard Deviation: Measures the spread of possible outcomes

(degree of risk).

• Decision Trees:
Graphical tools to analyze probabilistic outcomes and expected payoffs.

Example Calculation:
Alternative Outcome Probability Expected Value Contribution

A ₹100,000 0.6 ₹60,000

A ₹50,000 0.4 ₹20,000

Total EV(A) ₹80,000

Decision-maker chooses the option with maximum EV or adjusts for risk preferences.

5. Summary: Comparison of Decision Environments

Aspect Certainty Risk Uncertainty

Knowledge of Partial (probabilities None (probabilities

Complete
outcomes known) unknown)

Probabilistic /
Model type Deterministic Heuristic / Judgmental
Stochastic

Decision-making
Easy Moderate Difficult
difficulty

Information
High and accurate Moderate Low (but insufficient)
requirement

Linear programming, Decision trees, Maximin, Maximax,

Typical tools
cost models expected value Laplace criteria

Introducing new
Example Fixed deposit interest Stock investment
product

6. Key Takeaways

• Decision-making is influenced by the level of information about future

outcomes.

• Certainty: All outcomes known → deterministic decisions.

• Uncertainty: Outcomes unknown → rely on judgment or heuristics.

• Risk: Outcomes known with probabilities → probabilistic models.

 • Risk analysis uses expected value and probability-based reasoning to guide
optimal decisions.

• Most real-world decisions fall between risk and uncertainty rather than
complete certainty.

Lecture Notes: Decision Modeling with Spreadsheets

1. Introduction

Decision modeling is a vital component of prescriptive analytics, where models are

developed to recommend actions or solutions.
Among various implementation tools available today—such as programming languages
and modeling software—spreadsheets remain the most widely used platform for
Decision Support Systems (DSS) due to their flexibility, accessibility, and analytical
capabilities.

2. Role of Spreadsheets in Decision Modeling

2.1 Why Spreadsheets?

• Spreadsheets (e.g., Microsoft Excel, Google Sheets) are end-user modeling

tools used across business, engineering, mathematics, and science.

• They provide an interactive environment to build, test, and visualize models.

• Offer built-in functions for:

o Financial analysis

o Statistical analysis

o Forecasting

o Optimization

o Database management
3. Spreadsheet Add-ins for Modeling and DSS

Spreadsheets can be extended using add-ins that help perform specialized analytical
tasks.
These add-ins enhance the capability of spreadsheets for decision modeling.

Common Spreadsheet Add-ins

Add-in Developer / Source Purpose / Application

Solver Frontline Systems ([Link]) Linear and Nonlinear Optimization

What’sBest! Lindo Systems ([Link]) Optimization and DSS development

NeuralTools Palisade Corp. Artificial Neural Networks

Evolver Palisade Corp. Genetic Algorithms for optimization

@RISK Palisade Corp. Simulation and Risk Analysis

Note: Many comparable add-ins are available for free or at low cost and can be easily
searched online.

4. Advantages of Spreadsheets for DSS Development

Spreadsheets are popular because they:

1. Integrate multiple functions — financial, mathematical, and statistical tools in

one interface.

2. Support model solution tasks such as linear programming, regression, and

forecasting.

3. Allow rapid prototyping — easy to modify, visualize, and test models

interactively.

4. Provide transparency and traceability — all formulas and data are visible and
editable.

5. Enable “What-if” analysis, Goal Seeking, and Data management:

o What-if analysis: Change input values to see how output changes.

o Goal seeking: Specify desired output and let Excel find the required
input.

o Database management: Import or filter small data sets for analysis.

5. Real-World Applications of Spreadsheet-Based DSS

Case Study 1: Pennsylvania Adoption Exchange (PAE)

Problem:
PAE struggled to match children with adoptive families due to the vast database across
67 counties and inefficient algorithms.

Solution:
Developed a spreadsheet-based matching model where:

• Attributes of children and family preferences were entered.

• Caseworkers could assign weights to specific attributes (e.g., age, race,

location).

• The model computed match scores (0–100%) to find the best matches.

Results:

• Improved accuracy of family-child matches.

• Increased rate of successful adoptions.

• Reduced rejection rates by families and caseworkers.

Key Takeaway:
Spreadsheets can serve as effective, low-cost DSS tools even in nonprofit social
systems.

Case Study 2: Metro Meals on Wheels Treasure Valley (Idaho)

Problem:
Manual route planning for 800 meal deliveries was time-consuming and inefficient.

Solution:
Developed a spreadsheet-based routing tool using:

• Excel + VBA (Visual Basic for Applications).

• MapQuest API integration to compute:

o Distance and travel time between multiple stops.

o Optimal delivery routes (up to 30 stops).

• Automated validation of delivery addresses and route generation.

Results:

• Reduced travel distance by 10,000 miles/year.

• Saved 530 driving hours and $5,800 annually.

• Improved volunteer satisfaction and retention.

Intangible Benefits:

• Enhanced reliability of food delivery.

• Better resource utilization.

• Strengthened community service.

6. Spreadsheet Features Supporting Modeling

Feature Description

What-if Analysis Examines the effect of input changes on model outcomes.

Determines the required input to reach a specific target

Goal Seeking
output.

Macros /
Automates repetitive modeling tasks.
Programmability

Database Functions Import and manage small datasets or partial databases.

Templates and Reusable models and integration with other software (e.g.,
Integration databases, APIs).

7. Static vs. Dynamic Spreadsheet Models

Model Type Description Example

Static Represents a single snapshot or time Monthly loan payment calculation

Model period; does not change with time. for one period.

Represents behavior over time;

Dynamic Loan amortization model showing
variables evolve (e.g., payments,
Model prepayment effects over months.
balances).

Example:
 • A static loan model shows only one month’s calculation.

• A dynamic model tracks principal and interest across multiple periods and
incorporates prepayment behavior.

8. Risk Analysis and Simulation

• Spreadsheets can simulate uncertain conditions using random number

generators.

• Add-ins like @RISK or Crystal Ball perform Monte Carlo simulations directly in
Excel.

• These help in evaluating the risk and uncertainty in decision outcomes.

9. Integration and Flexibility

• Spreadsheets easily read/write common data formats (CSV, XML, etc.).

• Seamlessly integrate with databases, cloud platforms, and programming

tools.

• Excel’s interface resembles OLAP (Online Analytical Processing) dashboards,

enabling multidimensional analysis.

10. Summary

Spreadsheets have evolved from simple calculators to powerful analytical and

decision support systems.
They enable users to:

• Build and analyze both static and dynamic models.

• Conduct optimization, simulation, and risk analysis.

• Implement DSS solutions cost-effectively and intuitively.