Data Analytics Tutorial Solutions

Tutorial: Ride-Sharing Case Study (Slide 35)

Scenario: Reduce passenger wait times and improve route efficiency

using analytics.

Step 1: Data Collection & Preparation: Gather real-time and historical data from the
app, including GPS coordinates, ride timestamps, driver availability, traffic conditions,
and weather data. Clean the data to remove anomalies (e.g., GPS glitches).

Step 2: Descriptive Analytics (Analyze Patterns): Aggregate the historical data to

understand current patterns. Create heatmaps to identify high-demand zones
(hotspots) at specific times of day, and calculate the current average passenger wait
times across different geographic areas.

Step 3: Predictive Analytics (Predict Demand): Use time-series forecasting and

regression models (like ARIMA or Random Forest) on the historical data to forecast
future ride requests. This identifies exactly when and where demand will spike before it
happens.

Step 4: Prescriptive Analytics (Optimize Allocation): Implement optimization

algorithms to preemptively dispatch idle drivers to predicted high-demand areas.
Recommend the fastest routes to drivers in real-time, accounting for predicted traffic,
minimizing passenger wait times and maximizing driver utilization.
Tutorial: Making Sense of Data (Slide 36)

Describe each step in the process of making sense of data with a

suitable case study.

Step 1: Data Collection (Case: Retail Inventory): Gather raw data from Point of Sale
(POS) systems, e-commerce platforms, and warehouse logs detailing product sales,
dates, and stock levels over the past year.

Step 2: Data Cleaning & Preparation: Process the raw data by handling missing
values, removing duplicate transaction entries, and structuring the data into a unified,
analyzable format (e.g., a structured relational database).

Step 3: Data Analysis & Modeling: Apply statistical analysis and machine learning
models to identify trends, such as seasonal spikes in certain products or frequent item
pairings bought together by customers.

Step 4: Interpretation & Deployment: Translate the mathematical findings into

business strategy. In this case, adjusting the supply chain to stock more winter gear in
November and restructuring store aisles to place frequently paired items together.
How does data collected from IoT devices differ from data obtained
through surveys?

Step 1: Source and Nature: IoT data is collected passively and continuously from
machine sensors (e.g., temperature, GPS). Survey data is actively provided by humans
at discrete points in time.

Step 2: Volume and Velocity: IoT generates massive, high-velocity streams of big
data in real-time. Surveys produce much smaller, static datasets collected over days or
weeks.

Step 3: Objectivity vs. Subjectivity: IoT data is objective and quantitative, recording
exact physical states. Survey data is often subjective, capturing opinions, sentiments,
and self-reported behaviors which can be prone to bias.

Step 4: Structure and Application: IoT data is highly structured (time-series) and
used for operational monitoring or predictive maintenance. Survey data can be
unstructured (open text) and is typically used for market research and understanding
customer satisfaction.
 What are the advantages and disadvantages of using publicly available
datasets?

Step 1: Advantage - Accessibility & Cost: Public datasets (like those from Kaggle,
WHO, or government portals) are readily accessible and usually free, heavily reducing
the time and financial cost of data collection.

Step 2: Advantage - Benchmarking & Learning: They provide a standard baseline

for researchers and businesses to test new algorithms, benchmark performance
against others, and facilitate learning without privacy concerns.

Step 3: Disadvantage - Relevance & Specificity: Public datasets are generalized

and may not contain the specific variables or granularity required to solve a highly
niche, company-specific business problem.

Step 4: Disadvantage - Quality & Timeliness: There is no guarantee of data quality.

The data might be outdated, poorly maintained, contain hidden biases, or require
extensive cleaning before it can be trusted for analysis.

Tutorial: Descriptive Analytics (Slide 37)

What types of questions can descriptive analytics answer? Give at

least two examples.

Step 1: Focus on the Past: Descriptive analytics focuses strictly on summarizing

historical data to explain "What happened?" or "What is happening now?"

Step 2: Example 1 (Financial): "What was our total revenue and net profit margin for
the third quarter of this year compared to last year?"

Step 3: Example 2 (Operational): "Which of our three manufacturing plants produced

the highest number of defective units last month?"

Step 4: Value Generation: By answering these questions, organizations establish a

factual baseline of performance, usually visualized through dashboards, before moving
to more advanced predictive analytics.
Given a sales dataset, what descriptive statistics would you calculate
and why?

Step 1: Central Tendency (Mean, Median, Mode): I would calculate the mean and
median sales value per transaction to understand the average customer spend and
identify the most common purchase size (mode).

Step 2: Dispersion (Range, Variance, Standard Deviation): I would calculate the

standard deviation of daily sales to understand volatility—whether sales are consistent
day-to-day or highly erratic.

Step 3: Frequency Distribution: I would count the total number of items sold per
category and per region to see which products and locations are driving the most
volume.

Step 4: Percentiles & Quartiles: Calculating the 75th or 90th percentile helps identify
the threshold for "high-value" customers or top-performing days, useful for targeting VIP
segments.
Tutorial: Predictive Analytics (Slide 38)

Explain how historical data is used in predictive analytics.

Step 1: Data Accumulation: Predictive analytics requires a large repository of

historical data containing both the independent variables (features) and the known
outcomes (target variable).

Step 2: Pattern Extraction: Statistical models and machine learning algorithms scan
this historical data to identify underlying correlations, trends, and complex relationships
between the variables.

Step 3: Model Training: The algorithm uses the historical data as a "ground truth" to
train a predictive model. It learns the mathematical rules that lead from the input
features to the historical outcomes.

Step 4: Forecasting: Once trained and validated, the model processes new, unseen
current data to calculate probabilities and project what is most likely to happen in the
future.

Discuss a real-life application where predictive analytics has had a

significant impact such as healthcare and banking.

Step 1: Industry Selection (Banking): A major application of predictive analytics in

banking is Credit Scoring and Default Prediction.

Step 2: Historical Analysis: Banks analyze massive historical datasets containing

applicant demographics, income, past credit behavior, and the ultimate outcome
(whether they defaulted on a loan or paid it back).

Step 3: Predictive Modeling: Machine learning classification models (like Logistic

Regression or XGBoost) are trained on this data to output a "credit score" representing
the probability of a new applicant defaulting.

Step 4: Significant Impact: This automates loan approvals, drastically reduces the
financial risk of bad loans, ensures objective decision-making, and allows banks to offer
personalized interest rates based on risk levels.
Tutorial: Prescriptive Analytics (Slide 39)

What makes prescriptive analytics different from descriptive and

predictive analytics?

Step 1: Descriptive Role: Descriptive analytics looks backward, summarizing

historical data to answer "What happened?" (Hindsight).

Step 2: Predictive Role: Predictive analytics looks forward, using models to forecast
probabilities and answer "What will happen?" (Foresight).

Step 3: Prescriptive Role: Prescriptive analytics goes a step further by answering

"What should we do about it?". It recommends specific actions to achieve a desired
outcome.

Step 4: The Key Difference: While predictive stops at forecasting the future,
prescriptive analytics simulates various scenarios and applies optimization to actively
suggest the best decision path to maximize benefit or minimize risk.

Give an example where prescriptive analytics is used for decision

optimization.

Step 1: Example Scenario (Supply Chain Logistics): A global delivery company like
FedEx or UPS trying to optimize daily package delivery routes.

Step 2: Inputs & Predictions: The system takes in descriptive data (package
locations, truck capacities) and predictive data (expected traffic congestion, weather
forecasts for the day).

Step 3: Optimization Objective: The business defines mathematical constraints and

goals, such as minimizing total fuel consumption and ensuring all priority packages are
delivered before 10 AM.

Step 4: Prescriptive Output: The analytics engine calculates millions of permutations

and prescribes the exact turn-by-turn route each specific driver should take to meet all
objectives optimally.
What are some tools and techniques used in prescriptive analytics?

Step 1: Mathematical Optimization: Techniques like Linear Programming (LP),

Integer Programming, and Non-linear Programming are used to find the mathematical
maximum or minimum of an objective function subject to constraints.

Step 2: Simulation Techniques: Monte Carlo simulations and Discrete Event

Simulation (DES) are used to test thousands of "what-if" scenarios to observe potential
outcomes before making a decision.

Step 3: Advanced AI/Machine Learning: Reinforcement learning algorithms and

heuristic methods (like genetic algorithms) are increasingly used to prescribe actions in
highly complex, dynamic environments (e.g., self-driving cars).

Step 4: Software Tools: Industry-standard tools include IBM ILOG CPLEX, Gurobi
Optimizer, Microsoft Excel Solver, and Python libraries like SciPy, PuLP, and Pyomo.