UNIT 6

Applications of Data Science

Complete Question & Answer Reference

Subject Data Science & Applications

Unit 6 — Applications of Data Science

Total Questions 6 (Each carrying 10 Marks)

Total Marks 60

Table of Contents

Q1 Predictive Modeling — Definition, Working & Examples 10 Marks

Q2 Clustering & Customer Segmentation Application 10 Marks

Q3 Time Series Forecasting in Weather Prediction 10 Marks

Q4 Recommendation Engines & Product Recommendation 10 Marks

Q5 Four Applications of Data Science 10 Marks

Q6 Difference: Predictive Modeling vs Clustering 10 Marks

Page 1 | Unit 6 — Applications of Data Science

 Explain Predictive Modeling with an Example (House Price Prediction
Q1 10 Marks
or Fraud Detection)

1.1 Definition
Predictive modeling is a statistical and machine learning technique in which historical data is used to
build a mathematical model that can forecast future outcomes or unknown values. The model learns
patterns and relationships from past data and applies that knowledge to new, unseen data.

1.2 How Predictive Modeling Works

• Step 1 — Data Collection: Gather relevant historical data (e.g., past house sale records with
price, area, location).
• Step 2 — Data Preprocessing: Clean the data — handle missing values, remove duplicates, and
encode categorical variables.
• Step 3 — Feature Selection: Identify which variables (features) most influence the outcome.
• Step 4 — Model Training: Feed the data into a chosen algorithm (e.g., Linear Regression) so it
learns the mapping from inputs to outputs.
• Step 5 — Model Evaluation: Test the model on unseen data and measure performance using
metrics like accuracy or RMSE.
• Step 6 — Prediction: Deploy the trained model to predict outcomes for new data points.

1.3 Example A — House Price Prediction

A real-estate company wants to estimate the selling price of a house. It collects data from thousands of
past transactions. Each record contains features such as floor area (sq. ft.), number of bedrooms,
locality, age of the property, and distance from the nearest school or hospital. A Linear Regression
model is trained on this dataset. Once trained, the model can predict the price of any new house
instantly.

Example / Key Point: Input: Area = 1,200 sq. ft., Bedrooms = 3, Location = Pune. Predicted Output:
Selling Price = Rs. 72 Lakhs.

1.4 Example B — Fraud Detection in Banking

Banks process millions of transactions every day. A small fraction of these are fraudulent. A
classification model (e.g., Random Forest or Logistic Regression) is trained on labelled transaction data,
where each transaction is marked as 'Genuine' or 'Fraudulent'. The model learns patterns such as
unusual transaction amounts, foreign locations, or odd hours. When a new transaction is received, the
model immediately classifies it.

Example / Key Point: Transaction at 2:30 AM, Amount = Rs. 95,000, Location = Germany (account
holder is in Mumbai) → Model Output: FRAUD ALERT — Transaction Blocked.

1.5 Common Algorithms Used

Algorithm Best Used For

Page 2 | Unit 6 — Applications of Data Science

Linear Regression Predicting continuous values (e.g., house price)

Logistic Regression Binary classification (e.g., fraud / not fraud)

Decision Tree Rule-based predictions, easy to interpret

Random Forest High-accuracy classification and regression

Neural Networks Complex patterns in large datasets

Page 3 | Unit 6 — Applications of Data Science

 What is Clustering? Explain Customer Segmentation as an
Q2 10 Marks
Application of Clustering.

2.1 Definition of Clustering

Clustering is an unsupervised machine learning technique that automatically groups a set of data points
into clusters (subsets) such that points within the same cluster are highly similar to each other and
dissimilar to points in other clusters. Unlike predictive modeling, clustering does not require labelled
training data — the algorithm discovers the structure on its own.

2.2 Key Characteristics of Clustering

• Unsupervised: No predefined labels or categories are needed.
• Similarity-based: Data points are grouped by measuring distance or similarity.
• Exploratory: Used to discover unknown patterns in data.
• Popular Algorithms: K-Means, DBSCAN, Agglomerative Hierarchical Clustering.

2.3 Application — Customer Segmentation

Customer segmentation is the process of dividing a company's customer base into distinct groups
based on shared characteristics such as purchasing behavior, demographics, income level, or
frequency of interaction. It allows businesses to craft targeted marketing strategies and personalized
offers for each segment rather than applying a one-size-fits-all approach.

2.4 Step-by-Step Process (K-Means Clustering)

• Step 1: Collect customer data — age, annual income, purchase frequency, average order value.
• Step 2: Normalise the data so all features have equal weight.
• Step 3: Choose the number of clusters K (e.g., K = 4).
• Step 4: Run K-Means algorithm — it assigns each customer to the nearest cluster centroid.
• Step 5: Interpret each cluster and design marketing strategies accordingly.

2.5 Example — E-Commerce Customer Segments

Cluster Segment Name Characteristics Strategy

1 Premium Customers High income, frequent buyers, large orders Exclusive loyalty rewards, early access to products

2 Discount Seekers Buy only during sales, price-sensitive Flash sale alerts, coupon codes

3 New / Inactive Users Registered but rarely buy Welcome offers, tutorial emails, reminders

4 Regular Mid-Tier Consistent buyers, average spend Cross-sell related products, bundle offers

Example / Key Point: A retail chain using clustering found that 15% of its customers (Cluster 1)
generated 60% of total revenue. By targeting them with a VIP programme, revenue increased by 22%.

Page 4 | Unit 6 — Applications of Data Science

 Q3 How is Time Series Forecasting Used in Weather Forecasting? 10 Marks

3.1 What is a Time Series?

A time series is a sequence of numerical data points collected or recorded at successive,
equally-spaced points in time. Examples include daily temperature readings, hourly wind speed,
monthly rainfall totals, or annual average humidity levels.

3.2 What is Time Series Forecasting?

Time series forecasting is the technique of using historical time-stamped data to build a model that can
predict future values. It identifies underlying patterns in the data — such as trends, seasonal cycles, and
irregular fluctuations — and projects them forward.

3.3 Key Patterns in Time Series Data

Pattern Description Weather Example

Trend Long-term increase or decrease in valuesAverage temperature rising by 0.2°C per decade (global warming)

Seasonality Repeating pattern at a fixed interval Monsoon rainfall every June–September in India

Cyclical Irregular up-and-down fluctuations El Nino / La Nina effects every 3–7 years

Noise Random, unpredictable variation Sudden unseasonal rain or hailstorm

3.4 Application in Weather Forecasting

Meteorological organisations such as IMD (India Meteorological Department) and NOAA collect weather
measurements continuously from thousands of ground stations and satellites. This historical data,
spanning decades, is used to train time series models that forecast future weather conditions with high
accuracy.

3.5 Forecasting Workflow

• Data Collection: Automated weather stations record temperature, pressure, humidity, wind speed,
and rainfall every hour.
• Data Preprocessing: Handle missing sensor readings; smooth out measurement errors.
• Decomposition: Separate the time series into trend, seasonal, and residual (noise) components.
• Model Selection: Choose an appropriate algorithm (ARIMA, SARIMA, or LSTM neural network).
• Training: Fit the model on 20–50 years of historical weather data.
• Forecasting: Generate predictions for the next 1 to 14 days.
• Validation: Compare forecasts with actual observed values; refine the model regularly.

Example / Key Point: IMD uses 140 years of monsoon data to predict the onset of monsoon. Example
forecast: 'Mumbai will receive heavy rainfall (>115 mm) on 10 June 2025 — confidence: 87%'.

3.6 Algorithms Commonly Used

• ARIMA (AutoRegressive Integrated Moving Average) — for non-seasonal data.
• SARIMA (Seasonal ARIMA) — extends ARIMA to handle seasonal patterns like monsoon.

Page 5 | Unit 6 — Applications of Data Science

• Facebook Prophet — handles holidays, missing data, and multiple seasonalities easily.
• LSTM Neural Networks — captures long-range dependencies in complex weather patterns.

Page 6 | Unit 6 — Applications of Data Science

 What are Recommendation Engines? Explain with a Product
Q4 10 Marks
Recommendation Example.

4.1 Definition
A recommendation engine (also called a recommender system) is an intelligent data science system
that predicts and suggests items — products, movies, songs, articles — that a particular user is most
likely to find relevant or interesting. It analyses user behaviour, preferences, and similarities with other
users to generate personalised recommendations.

4.2 Types of Recommendation Engines

Type How It Works Example

Collaborative Recommends items that users with similar behaviour liked.

'Users like you also bought X' (Amazon)
Filtering

Content-Based Recommends items similar to what the user previously interacted

'Becausewith.
you watched Action Movie A, try B' (Netflix)
Filtering

Hybrid Combines collaborative and content-based methods for better

Netflix,
accuracy.
Spotify, and Amazon all use hybrid models.
Filtering

Knowledge-Based Uses explicit user preferences and domain rules. Financial product advisors, travel booking sites.

4.3 Example — Amazon Product Recommendation

Amazon operates one of the most advanced recommendation engines in the world. When a customer
visits the platform, the system analyses multiple signals to suggest products.

• Signal 1 — Browsing History: Products the user recently viewed.

• Signal 2 — Purchase History: Items the user has bought in the past.
• Signal 3 — Ratings & Reviews: Products the user has rated highly.
• Signal 4 — Similar Users: What other users with a matching profile purchased.
• Signal 5 — Item Similarity: Products that are frequently bought together.

Example / Key Point: A user searches for 'Sony WH-1000XM5 Wireless Headphones'. The engine
recommends: JBL Portable Bluetooth Speaker | USB-C Charging Cable | Laptop Stand | Microfibre
Cleaning Cloth. This increases Amazon's average order value by 35%.

4.4 Real-World Applications

• Netflix: Recommends movies and series based on watch history; 80% of viewing comes from
recommendations.
• Spotify: Generates personalised playlists (Discover Weekly) using listening history.
• YouTube: Suggests 'Up Next' videos; accounts for 70% of total watch time.
• Swiggy / Zomato: Recommends restaurants and dishes based on past orders and location.
• LinkedIn: Suggests jobs, connections, and courses based on profile and behaviour.

Page 7 | Unit 6 — Applications of Data Science

 Q5 List and Explain any Four Applications of Data Science. 10 Marks

Data science is an interdisciplinary field that combines statistics, programming, and domain knowledge
to extract actionable insights from data. It is applied extensively across industries. Four important
applications are described below.

5.1 Healthcare — Disease Prediction and Medical Diagnosis

Data science enables early and accurate detection of diseases by analysing patient records, medical
images, genetic data, and clinical notes. Machine learning models are trained on large datasets of
diagnosed patients and can detect conditions that even experienced clinicians might miss at early
stages.

Example / Key Point: Google DeepMind's AI detects over 50 eye diseases from retinal scans with
accuracy comparable to specialist ophthalmologists. IBM Watson assists oncologists in recommending
cancer treatment plans by analysing millions of medical publications.

5.2 Finance — Fraud Detection and Credit Risk Assessment

Financial institutions process millions of transactions daily. Data science models monitor every
transaction in real time, comparing it against historical patterns to flag anomalies. Additionally, credit
scoring models assess the likelihood that a loan applicant will default, enabling banks to make informed
lending decisions.

Example / Key Point: HDFC Bank's fraud detection system blocks suspicious transactions within 200
milliseconds. CIBIL score in India is computed using a predictive model trained on millions of credit
histories.

5.3 E-Commerce and Retail — Recommendation and Demand Forecasting

E-commerce platforms use recommendation engines to personalise the shopping experience,
increasing both customer satisfaction and revenue. Simultaneously, demand forecasting models predict
how much of each product to stock at each location, reducing waste and improving supply-chain
efficiency.

Example / Key Point: Amazon's recommendation engine contributes to approximately 35% of its total
annual revenue. Walmart uses data science to forecast product demand and manage inventory across
10,000+ stores.

5.4 Transportation — Route Optimisation and Predictive Maintenance

Ride-hailing and logistics companies use data science to minimise delivery times and costs by finding
optimal routes in real time. Airlines and railways use predictive maintenance models to detect potential
equipment failures before they occur, preventing costly breakdowns and improving passenger safety.

Page 8 | Unit 6 — Applications of Data Science

Example / Key Point: Uber's ML model dynamically adjusts surge pricing and driver allocation based on
real-time demand. Indian Railways uses sensor data and ML to predict wheel and track defects 2–3
weeks in advance.

Page 9 | Unit 6 — Applications of Data Science

 Differentiate between Predictive Modeling and Clustering with
Q6 10 Marks
Examples.

Predictive modeling and clustering are both core techniques in data science but serve fundamentally
different purposes, use different types of data, and produce different kinds of outputs.

6.1 Comparison Table

Parameter Predictive Modeling Clustering

Learning Type Supervised Learning Unsupervised Learning

Labelled Data Required — historical outcomes must be known

Not required — labels are absent

Primary Goal Predict a specific value or category for new data

Discover hidden groups or patterns in data

Output A predicted value (e.g., price) or class (e.g., fraud/no

A set offraud)
clusters; each data point assigned to a group

Algorithms Linear Regression, Logistic Regression, SVM,K-Means,

Random Forest
DBSCAN, Agglomerative Clustering

Evaluation Metrics Accuracy, Precision, Recall, RMSE, F1-Score Silhouette Score, Davies-Bouldin Index, Inertia

When to Use When the outcome variable is known and prediction

When isstructure
the goalin data is unknown and needs to be explored

Example Predicting whether a patient has diabetes (YesGrouping

/ No) patients by symptoms without prior diagnosis

Data Requirement Large labelled dataset needed for training Works even without any labels; larger data improves quality

6.2 Illustrative Example — Banking Scenario

The following example uses a banking context to clearly show when each technique is applied:

Predictive Modeling Scenario:

A bank wants to decide whether to approve or reject a new loan application. It has 10 years of past loan
records where each application is labelled 'Repaid' or 'Defaulted'. A Logistic Regression model is
trained on this data using features such as monthly income, existing debt, credit score, and employment
status. When a new application arrives, the model predicts the probability of default.

Example / Key Point: Input: Income = Rs. 45,000/month, Credit Score = 680, Existing Loans = 2. Model
Output: Probability of Default = 34% → Loan APPROVED with higher interest rate.

Clustering Scenario:
The same bank wants to understand its existing customer base without any predefined categories. It
runs a K-Means clustering algorithm on customer transaction data. The algorithm automatically
discovers four distinct groups without any labels being provided.

Cluster Profile Discovered Action Taken by Bank

Cluster A Young, high-spending, frequent digital transactions Offer premium credit cards and investment apps

Cluster B Middle-aged, moderate income, regular savings Promote fixed deposits and home loan schemes

Page 10 | Unit 6 — Applications of Data Science

Cluster C Senior citizens, low transactions, high savings Offer senior citizen savings schemes and insurance

Cluster D Young students, low income, mobile-first users Promote student accounts and UPI cashback offers

6.3 Summary — When to Choose Which Technique

• Use Predictive Modeling when you have a specific question with a known answer type and
labelled historical data available (e.g., 'Will this customer churn?').
• Use Clustering when you want to explore your data without prior assumptions and discover
natural groupings that were previously unknown (e.g., 'What types of customers do we have?').
• Use Both Together: Clustering can generate labels (customer segments) that are then used as
inputs to a predictive model for even richer analysis.

Unit 6 — Applications of Data Science | 6 Questions | Total: 60 Marks

Page 11 | Unit 6 — Applications of Data Science