MODULE III: TYPES OF ANALYTICS

Curricular Scope & Pedagogical Objective: Module III delivers a comprehensive, mathematically rigorous, and managerially grounded exploration of analytical classifications within the Calicut University B.Com (Honours) curriculum for Data Analytics for Business (COM2MN109). It investigates the conceptual transition from data to information in decision science; contrasts Traditional Reporting with Gartner's four-tier analytical hierarchy (Descriptive, Diagnostic, Predictive, and Prescriptive Analytics); details predictive statistical techniques (Linear and Logistic Regression, Time-Series forecasting, Machine Learning classification) and model evaluation metrics (Confusion Matrix, Precision, Recall, F1-Score); evaluates prescriptive optimization algorithms (Linear Programming, Monte Carlo simulation); examines Big Data architectures (Hadoop, Spark, Kafka, NoSQL); and articulates cross-functional applications across marketing, financial risk scoring, supply chain operations, and human resource management.

Unit 10: Data vs. Information & Data and Decision Making

10.1 Conceptual Transition: From Raw Data to Contextualized Information

In information economics and decision theory, data and information represent distinct stages in the cognitive value chain:

Data: Raw, uncontextualized, unorganized facts, measurements, symbols, or transaction records. In isolation, raw data carries zero inherent meaning and cannot guide organizational behavior (e.g., an isolated number: "250").
Information: Data that has been processed, structured, formatted, and contextualized to confer relevance and purpose. Information resolves uncertainty and answers fundamental operational questions: Who? What? Where? When? (e.g., "The retail store sold 250 units of air conditioners in Palakkad during April, representing a 40% year-on-year increase").

10.2 The Economics of Information in Business Decision Theory

In microeconomic decision theory, perfect information eliminates decision uncertainty. However, in real-world commercial markets, decision-makers operate under Bounded Rationality (pioneered by Nobel Laureate Herbert Simon). Managers face computational constraints, cognitive limits, and information asymmetry.

Data analytics serves as the economic mechanism that reduces Information Asymmetry—the condition wherein one party to a commercial transaction possesses superior information compared to the other (e.g., a borrower knowing their true default risk while the lending bank is uninformed). By deploying predictive credit scoring models, financial institutions neutralize information asymmetry, slashing loan default rates and lowering systemic interest costs.

10.3 Establishing the Single Source of Truth (SSOT)

A critical failure in traditional enterprise decision-making occurs when competing organizational departments present conflicting figures for the exact same business metric (e.g., the Sales team reporting ₹50 Crores in revenue based on booked contracts, while the Finance team reports ₹42 Crores based on realized bank receipts). Modern data analytics establishes a Single Source of Truth (SSOT) through unified enterprise data warehouses, centralized semantic data models, and Master Data Management (MDM), ensuring executive decisions are anchored in synchronized, indisputable facts.

Unit 11: Types of Analytics: Descriptive, Predictive, Prescriptive & Traditional

11.1 Traditional Analytics / Historical Reporting

Traditional Analytics (often termed static Management Information Systems or MIS reporting) represents the historical baseline of corporate data usage. Characterized by periodic batch generation (monthly, quarterly), it produces static tabular financial ledgers, inventory balance sheets, and sales spreadsheets.

Inherent Limitations of Traditional Reporting:
• Extreme Latency: By the time a quarterly printed report reaches the CEO's desk, the market conditions have already shifted.
• Lack of Interactivity: Static reports cannot be filtered, drilled down, or queried on the fly.
• Inability to Explain Causation: Shows that revenues dropped, but cannot explain why or project what happens next.

11.2 Descriptive Analytics: "What Happened?"

Descriptive Analytics applies mathematical aggregation and statistical summarization to historical data to describe operational reality and establish performance baselines:

1. Measures of Central Tendency & Dispersion

• Central Tendency: Arithmetic Mean (average), Median (middle value robust to outliers), and Mode (most frequent category).
• Dispersion & Spread: Variance, Standard Deviation (σ), Range, and Interquartile Range (IQR) measuring operational volatility.
• Shape: Skewness (asymmetry) and Kurtosis (tail heaviness / outlier propensity).

2. Multidimensional Data Visualization

Transforms numerical tables into cognitive visual representations: interactive bar charts, time-series line trends, scatterplots, geospatial choropleth maps, and real-time executive KPI dashboards in Microsoft Power BI and Tableau.

11.3 Diagnostic Analytics: "Why Did It Happen?"

Diagnostic Analytics bridges historical observation and future prediction by drilling down into datasets to isolate root causes, anomalies, and underlying drivers of observed outcomes:

Root-Cause Analysis (RCA): Investigating multi-dimensional data filters to discover why an unexpected spike or plunge occurred (e.g., discovering that a sudden drop in regional sales was confined to a specific delivery courier's strike).
Variance Analysis: Comparing actual financial results against budgeted benchmarks and isolating price variance from volume variance.
Correlation Matrices: Calculating Pearson (r) and Spearman (ρ) correlation coefficients to evaluate the strength and direction of linear relationships between business variables.

11.4 Predictive Analytics: "What Will Happen?"

Predictive Analytics utilizes historical data patterns, statistical modeling, machine learning algorithms, and probability theory to forecast future events, customer behaviors, and commercial trends:

PREDICTIVE MODELING ALGORITHMS STATISTICAL & ML ARCHITECTURE

Y = eta_0 + eta_1 X_1 + dots + eta_k X_k + epsilon, quad P(Y=1) = rac{1}{1 + e^{-(eta_0 + eta_1 X)}}

Core Predictive Methodologies:

Multiple Linear Regression: Models continuous numerical outcomes (e.g., forecasting sales revenue based on advertising spend, store square footage, and price).
Logistic Regression: Models binary categorical probabilities (between 0 and 1) via the sigmoid/logit link function (e.g., predicting customer churn: Yes/No, or loan default: Yes/No).
Time-Series Forecasting: Models temporal sequence data incorporating trend, seasonality, and cyclic fluctuations using Moving Averages, Exponential Smoothing, and ARIMA (Autoregressive Integrated Moving Average) models.
Machine Learning Classification Trees: Decision Trees, Random Forests, and Gradient Boosting (XGBoost) that split data along non-linear decision boundaries to classify risk.

Evaluating Predictive Classification Models: The Confusion Matrix

In classification modeling, accuracy alone is misleading (e.g., in fraud detection where 99.9% of transactions are legitimate, a model that simply predicts "No Fraud" for every transaction achieves 99.9% accuracy while catching zero fraud). Data analysts evaluate models using a Confusion Matrix:

Confusion Matrix	Predicted Negative (0)	Predicted Positive (1)
Actual Negative (0)	True Negative (TN): Correctly classified non-fraud.	False Positive (FP / Type I Error): False alarm; legitimate transaction blocked.
Actual Positive (1)	False Negative (FN / Type II Error): Missed fraud; catastrophic loss.	True Positive (TP): Correctly intercepted fraud.

PREDICTIVE EVALUATION METRICS MATHEMATICAL DEFINITIONS

ext{Accuracy} = rac{TP + TN}{ ext{Total}}, quad ext{Precision} = rac{TP}{TP + FP}, quad ext{Recall} = rac{TP}{TP + FN}, quad F_1 = 2 imes rac{ ext{Precision} imes ext{Recall}}{ ext{Precision} + ext{Recall}}

Precision: Out of all transactions predicted as fraud, how many were actually fraudulent? (Minimizes customer irritation from false alarms).
Recall (Sensitivity): Out of all actual fraudulent transactions, how many did the model intercept? (Minimizes financial losses from missed fraud).
F1-Score: Harmonic mean of Precision and Recall; provides a balanced metric for imbalanced datasets.

11.5 Prescriptive Analytics: "What Should We Do?"

Prescriptive Analytics represents the pinnacle of analytical maturity. It does not merely forecast what will happen; it recommends the specific course of action to achieve an optimal commercial outcome, evaluating trade-offs, operating constraints, and risk profiles:

1. Mathematical Optimization (Linear Programming)

Uses algorithms (e.g., the Simplex algorithm) to maximize an objective function (e.g., profit) or minimize a cost function subject to scarce resource constraints (machine hours, labor limits, raw material availability).

2. Monte Carlo Simulation

Executes tens of thousands of probabilistic simulations, randomly sampling variable distributions (sales growth, exchange rates, raw material costs) to produce a probability distribution of net present value (NPV) and project risk.

Unit 12: Big Data Analytics & Technological Architectures

12.1 The Big Data Technology Stack

Processing Petabyte-scale enterprise data requires distributed, fault-tolerant computing architectures:

Apache Hadoop Ecosystem:
- HDFS (Hadoop Distributed File System): Splits massive datasets into blocks (e.g., 128MB) and distributes them redundantly across clusters of commodity hardware servers.
- MapReduce: A parallel programming model that breaks computational queries into a "Map" phase (filtering/sorting) and a "Reduce" phase (aggregating summary results).
Apache Spark: An open-source unified analytics engine that executes computations in-memory (RAM) rather than reading/writing to physical disk, running queries up to 100 times faster than legacy MapReduce.
Streaming & Event-Driven Architectures (Apache Kafka & Flink): Enables continuous, real-time data streaming pipelines, ingesting millions of clickstream events or stock quotes per second with sub-second latency.
NoSQL Multi-Model Stores: Key-Value (Redis), Document (MongoDB), Wide-Column (Apache Cassandra), and Graph databases (Neo4j) that scale horizontally without relational joins.

12.2 Text Analytics & Natural Language Processing (NLP)

Over 80% of corporate data resides in unstructured text (emails, social media reviews, customer support chats, legal contracts). Text Analytics uses computational linguistics and NLP to extract structured commercial intelligence:

1. Sentiment Analysis (Opinion Mining)

Classifies textual reviews into positive, neutral, or negative polarities, scoring emotional intensity to monitor brand reputation in real time.

2. Topic Modeling & Entity Extraction

Unsupervised algorithms (e.g., Latent Dirichlet Allocation - LDA) that automatically cluster thousands of customer complaints into thematic buckets (e.g., "billing error", "packaging defect", "delivery delay").

Unit 13: Cross-Functional Applications of Data Analytics in Business

13.1 Marketing & Customer Analytics

Marketing analytics optimizes customer acquisition, retention, and basket size:

MARKET BASKET ANALYSIS: THE APRIORI METRICS ASSOCIATION RULE MINING

ext{Support}(A o B) = rac{ ext{Count}(A cap B)}{N}, quad ext{Confidence}(A o B) = rac{P(A cap B)}{P(A)}, quad ext{Lift} = rac{ ext{Confidence}}{P(B)}

Association Rule Interpretation:

Support: How frequently the combination of items appears in all transactions.
Confidence: If a customer purchases Item A (e.g., Bread), what is the conditional probability they also purchase Item B (e.g., Butter)?
Lift Ratio: Measures the strength of the association rule over random chance. A Lift > 1.0 indicates that the purchase of Item A significantly increases the likelihood of purchasing Item B (used by supermarkets to design shelf layouts and e-commerce cross-sell banners).

13.2 Financial & Credit Analytics

In commercial banking and investment management:

Credit Risk Scoring: Estimating probability of default ( $P D$ ) using logistic regression and decision trees, setting risk-adjusted interest rates.
Algorithmic Trading & Portfolio Optimization: Deploying Markowitz Modern Portfolio Theory and Sharpe Ratio optimization algorithms to balance risk-return frontiers.

13.3 Supply Chain & Operations Analytics

In manufacturing and retail distribution:

Dynamic Inventory Optimization: Calculating safety stocks and reorder points under stochastic demand to eliminate stockouts while minimizing holding costs.
Predictive Maintenance: IoT sensor telemetry predicting machine bearing failure 50 hours in advance, avoiding factory line downtime.

13.4 Human Resource (HR) & People Analytics

In human capital management:

Employee Flight Risk Modeling: Identifying key personnel likely to resign based on tenure, promotion latency, and compensation equity, enabling proactive retention.
Workforce Shift Scheduling: Balancing staff allocations against hourly retail footfall trends.

Analytical Application: Quantitative Exercises & Calculations

PRACTICAL PROBLEM 1: EVALUATING PREDICTIVE FRAUD CLASSIFICATION
Case Scenario: A FinTech bank tests a machine learning fraud detection model on a validation dataset of 10,000 credit card transactions. Historical forensic auditing reveals that out of the 10,000 transactions, exactly 200 were fraudulent (Positive) and 9,800 were legitimate (Negative).
The model's classification output is as follows:
• True Positives (TP): 160 fraudulent transactions correctly flagged.
• False Positives (FP): 40 legitimate transactions falsely blocked (False Alarms).
• False Negatives (FN): 40 fraudulent transactions missed by the model.
• True Negatives (TN): 9,760 legitimate transactions correctly approved.
Calculate: (a) Model Accuracy, (b) Precision, (c) Recall (Sensitivity), and (d) F1-Score.

Step 1: Calculate Overall Accuracy:
Accuracy = (TP + TN) ÷ Total Transactions
Accuracy = (160 + 9,760) ÷ 10,000 = 9,920 ÷ 10,000 = 99.20%

Step 2: Calculate Precision (Positive Predictive Value):
Precision = TP ÷ (TP + FP) = 160 ÷ (160 + 40) = 160 ÷ 200 = 0.80 (80.0%)
• When the model flags a transaction as fraud, it is correct 80% of the time.

Step 3: Calculate Recall (Sensitivity):
Recall = TP ÷ (TP + FN) = 160 ÷ (160 + 40) = 160 ÷ 200 = 0.80 (80.0%)
• The model successfully intercepts 80% of all actual fraudulent transactions.

Step 4: Calculate F1-Score:
F1-Score = 2 × (Precision × Recall) ÷ (Precision + Recall)
F1-Score = 2 × (0.80 × 0.80) ÷ (0.80 + 0.80) = 2 × 0.64 ÷ 1.60 = 1.28 ÷ 1.60 = 0.80

PRACTICAL PROBLEM 2: PRESCRIPTIVE ANALYTICS VIA LINEAR PROGRAMMING (LP)
Case Scenario: A manufacturing firm produces two commercial products: Standard (X₁) and Deluxe (X₂).
• Profit Contribution: ₹40 per unit of X₁; ₹60 per unit of X₂.
• Machine Time Constraint: X₁ requires 2 hours; X₂ requires 4 hours. Total available machine time = 120 hours.
• Labor Time Constraint: X₁ requires 3 hours; X₂ requires 2 hours. Total available labor time = 90 hours.
Formulate the complete Linear Programming problem and identify the optimal product mix using corner-point graphical method.

Step 1: Mathematical Formulation:
Maximize Z (Profit) = 40X₁ + 60X₂
Subject to:
(1) Machine Constraint: 2X₁ + 4X₂ ≤ 120
(2) Labor Constraint: 3X₁ + 2X₂ ≤ 90
(3) Non-Negativity: X₁ ≥ 0, X₂ ≥ 0

Step 2: Solve Corner Points of Feasible Region:
• Point A (Origin): (0, 0) → Z = ₹0
• Point B (Y-intercept of Machine constraint): Set X₁ = 0 in (1): 4X₂ = 120 → X₂ = 30 → (0, 30)
• Point C (X-intercept of Labor constraint): Set X₂ = 0 in (2): 3X₁ = 90 → X₁ = 30 → (30, 0)
• Point D (Intersection of both constraints):
Multiply (2) by 2: 6X₁ + 4X₂ = 180
Subtract (1): (6X₁ + 4X₂) − (2X₁ + 4X₂) = 180 − 120 → 4X₁ = 60 → X₁ = 15
Substitute into (1): 2(15) + 4X₂ = 120 → 30 + 4X₂ = 120 → 4X₂ = 90 → X₂ = 22.5 units

Step 3: Evaluate Profit (Z) at All Feasible Corner Points:
• Point A (0, 0): Z = 40(0) + 60(0) = ₹0
• Point B (0, 30): Z = 40(0) + 60(30) = ₹1,800
• Point C (30, 0): Z = 40(30) + 60(0) = ₹1,200
• Point D (15, 22.5): Z = 40(15) + 60(22.5) = 600 + 1,350 = ₹1,950 (MAXIMUM PROFIT)
• Prescriptive Decision Recommendation: The enterprise should manufacture exactly 15 units of Standard and 22.5 units of Deluxe to achieve maximum profit of ₹1,950.

Real-World Empirical Corporate Case Studies

CASE STUDY 1: Zara (Inditex) — Real-Time Descriptive and Prescriptive Fast-Fashion Analytics

While traditional apparel fashion brands plan collections 9 months in advance, Spanish retail giant Inditex (Zara) delivers new fashion designs to retail stores worldwide in under 15 days.

Analytical Feedback Loop: Store managers use handheld devices to record real-time descriptive customer feedback (e.g., requests for altered collar cuts or colors). This data streams directly to designers at headquarters in Arteixo, Spain. Prescriptive optimization algorithms analyze fabric inventory, factory capacity, and store sales velocities to automatically generate small production batches, slashing unsold inventory markdowns to under 15% (compared to 30-40% across the global fashion industry).

CASE STUDY 2: Uber — Dynamic Surge Pricing via Geospatial Prescriptive Analytics

Ride-hailing giant Uber processes billions of GPS data pings daily. Uber's dynamic pricing algorithm is a premier real-world manifestation of prescriptive analytics.

Algorithmic Equilibrium: When localized passenger ride requests spike in a specific urban neighborhood (e.g., following a concert or sudden rainstorm), demand exceeds available driver supply. Prescriptive algorithms calculate a dynamic surge multiplier, raising ride prices in real time. This surge serves two simultaneous economic functions: it dampens excess consumer demand while incentivizing distant drivers to enter the surge zone, rapidly restoring supply-demand equilibrium.

CASE STUDY 3: American Express — Real-Time Predictive Machine Learning for Fraud Prevention

American Express processes over $1 Trillion in transactions annually. To protect cardholders without causing false transaction denials, AmEx deployed predictive machine learning algorithms operating on big data clusters.

Millisecond Scoring: In less than 10 milliseconds, the predictive engine scores every card swipe across hundreds of features: merchant category, cardholder purchasing history, device telemetry, and time of day. This predictive interception prevents over $1 Billion in fraudulent charges annually while maintaining the lowest fraud loss rate in the payments industry.

Exam Preparation & University Review Questions

CALICUT UNIVERSITY B.COM (FYUGP) HIGH-FREQUENCY EXAMINATION QUESTIONS

Part A: Short-Answer Questions (Conceptual - 2 Marks Each)

Differentiate between "Data" and "Information".
What is a "Single Source of Truth" (SSOT) in data management?
Define "Descriptive Analytics" and name two common descriptive metrics.
What question does "Diagnostic Analytics" seek to answer?
Define "Predictive Analytics" with one commercial banking example.
What is a "Confusion Matrix"?
Define "Prescriptive Analytics".
What is the "Lift Ratio" in Market Basket Analysis?

Part B: Short-Essay / Analytical Questions (5 Marks Each)

Examine the key limitations of traditional historical MIS reporting compared to modern Business Analytics.
Explain the four tiers of Gartner's Analytic Ascendancy framework (Descriptive, Diagnostic, Predictive, Prescriptive) with business illustrations.
Detail the components of a Confusion Matrix and define Accuracy, Precision, Recall, and F1-Score.
Describe the core technologies comprising the Big Data ecosystem (Hadoop HDFS, Spark, Kafka, NoSQL).
How does Market Basket Analysis (Apriori algorithm) utilize Support, Confidence, and Lift to optimize retail merchandise placement?

Part C: Comprehensive Essay Questions (10 / 15 Marks Each)

"Analytics has evolved from retrospective bookkeeping to autonomous decision guidance." Critically examine this statement. Detail the concepts, techniques, mathematical tools, and commercial applications of Descriptive, Predictive, and Prescriptive Analytics with real-world examples.
Explain the mathematical formulation and graphical solution of Linear Programming in Prescriptive Analytics. Using a hypothetical production scenario, demonstrate how an enterprise optimizes resource allocation subject to constraints.
Critically evaluate the cross-functional applications of Data Analytics in Marketing, Financial Services, Supply Chain Operations, and Human Resources. Discuss how organizations like Zara and Uber achieve market supremacy through advanced analytics.

Key Takeaways for Examination Distinction:

Clearly state the core question answered by each tier: Descriptive (What?), Diagnostic (Why?), Predictive (What will happen?), Prescriptive (What should we do?).
In Confusion Matrix questions, always draw the 2x2 table labeling True Positives, False Positives (Type I Error), False Negatives (Type II Error), and True Negatives. Write out the exact F1-Score formula.
For Linear Programming problems, clearly write the three required elements: Objective Function, Structural Constraints, and Non-Negativity Constraints.
Quote Zara's 15-day fast-fashion loop and Uber's surge pricing as benchmark case studies in prescriptive analytics.