the first batch
(Q1–Q5):
Q1. Explain the concept of analytics, its importance in
today's business environment, and discuss how it aids in strategic
decision-making with relevant examples.
Answer:
Analytics refers to the systematic analysis of data to uncover
meaningful patterns and insights that support decision-making. It plays a
central role in helping organizations understand past performance, predict
future outcomes, and make informed decisions. There are three primary types of
analytics: descriptive (what happened), predictive (what will happen), and
prescriptive (what to do next).
Importance:
In the digital age, companies generate vast amounts of data
through customer interactions, operational systems, and third-party sources.
Analytics enables organizations to derive value from this data. For example:
* Retailers use customer data to forecast demand and
personalize marketing.
* Financial institutions detect fraud through anomaly
detection.
* Healthcare providers predict patient readmission risks.
Strategic Decision-Making:
Analytics supports strategy by revealing trends, identifying
risks and opportunities, and testing hypotheses. For example, Starbucks uses
location analytics to determine store placement, and Netflix uses predictive
analytics to recommend content, enhancing user satisfaction.
Q2. Analyze the major challenges faced by organizations in
implementing analytics solutions and suggest approaches to overcome these
challenges.
Answer:
Challenges include:
1. Data Quality Issues: Inaccurate, inconsistent, or
incomplete data reduces model effectiveness.
2. Skills Gap: Lack of data science professionals and domain
experts.
3. Organizational Resistance: Employees may resist
data-driven changes.
4. Integration with Legacy Systems: Old IT infrastructure
often lacks compatibility with modern tools.
5. Cost and ROI Uncertainty: High upfront costs and unclear
long-term returns.
6. Ethical and Privacy Concerns: Compliance with regulations
like GDPR.
Approaches:
* Establish robust data governance and data quality
programs.
* Invest in employee training and attract skilled
professionals.
* Promote a culture of data-driven decision-making.
* Use cloud-based, scalable platforms to reduce
infrastructure burden.
* Start with small, high-impact projects to demonstrate ROI.
* Maintain transparency and follow ethical guidelines when
handling data.
Q3. Discuss the various applications of predictive analytics
across different industry sectors. Provide detailed case studies to illustrate
successful implementations.
Answer:
Applications:
* Retail: Demand forecasting, customer segmentation.
* Finance: Credit scoring, fraud detection.
* Healthcare: Disease prediction, hospital readmission.
* Manufacturing: Predictive maintenance, quality assurance.
* Telecom: Customer churn prediction.
Case Studies:
1. Target Corporation: Used predictive analytics to identify
pregnant customers based on purchase behavior, enabling personalized marketing
and early customer retention.
2. General Electric (GE): Integrated sensors on jet engines
for predictive maintenance, reducing downtime and saving millions.
3. Netflix: Uses viewing patterns to recommend shows,
improving viewer retention and engagement.
Q4. Demonstrate the evolution of analytics as a field and
explain why it has gained such popularity among businesses in the past decade.
Include industry experts' perspectives in your answer.
Answer:
Evolution:
* 1990s: Descriptive analytics using spreadsheets and basic
BI tools.
* 2000s: Diagnostic analytics with OLAP and data mining.
* 2010s: Predictive analytics using machine learning.
* 2020s: Prescriptive analytics with AI and automation.
Reasons for Growth:
* Explosion of big data and IoT.
* Affordable cloud computing.
* Advances in ML/AI algorithms.
* Competitive pressure and demand for innovation.
Industry Perspectives:
* Thomas Davenport emphasized "competing on analytics" as a
business imperative.
* Gartner identified data and analytics as top technology
priorities for CEOs.
Examples:
* Uber uses real-time predictive analytics for dynamic
pricing and routing.
* Amazon leverages data to optimize supply chain and product
recommendations.
Q5. Explain with a detailed analysis of a business case
where predictive analytics transformed operations. Discuss the problems and
approach.
Answer:
Case Study: UPS (United Parcel Service)
Problem:
UPS faced issues with inefficient routes, high fuel
consumption, and unexpected vehicle breakdowns.
Approach:
* Implemented predictive analytics using telematics sensors
in trucks.
* Built models to forecast mechanical issues (predictive
maintenance).
* Optimized delivery routes using ORION (On-Road Integrated
Optimization and Navigation).
Results:
* Saved millions of gallons of fuel.
* Reduced CO2 emissions and improved delivery punctuality.
* Enhanced operational efficiency and customer satisfaction.
Predictive analytics empowered UPS to transition from
reactive to proactive operations.
Next: I will send Q6–Q10 in the following message.
Here is the next batch of full answers: Questions 6–10 from
the CSE3036 – Predictive Analytics question bank.
Q6. Explain the ethical considerations and privacy concerns
related to the use of predictive analytics. How can organizations address these
issues?
Answer:
Ethical Considerations:
* Bias and Discrimination: Algorithms may inherit historical
bias in data, leading to unfair treatment of certain groups. For example,
predictive hiring tools may unintentionally discriminate against minorities.
* Lack of Transparency: Complex models like deep learning
may be difficult to interpret, making it unclear why decisions are made.
* Manipulation: Predictive models can be used to nudge users
unethically (e.g., manipulating purchase decisions).
* Consent and Autonomy: Users may not be aware that their
data is being used to predict or influence their behavior.
Privacy Concerns:
* Data Collection: Massive data collection from mobile apps,
sensors, and social platforms can infringe on individual privacy.
* Data Sharing: Sharing customer data with third parties can
lead to misuse or unauthorized access.
* Data Breaches: Predictive systems are vulnerable to
cyberattacks, risking exposure of sensitive information.
* Re-identification: Even anonymized datasets can sometimes
be reverse-engineered to identify individuals.
Solutions:
* Implement strong data governance and compliance with
privacy regulations like GDPR and HIPAA.
* Adopt transparent models and use Explainable AI (XAI) to
improve understanding.
* Limit data collection to only what is necessary (data
minimization).
* Regularly audit models for fairness and bias.
* Clearly inform users about data usage and obtain consent.
Q7. Compare and contrast the different types of analytics
with examples of how each contributes to business value.
Answer:
There are four main types of analytics:
1.
Descriptive Analytics:
* What happened?
* Uses historical data to summarize outcomes.
* Example: Monthly sales report.
* Business Value: Offers situational awareness and
reporting.
2.
Diagnostic Analytics:
* Why did it happen?
* Investigates the root cause of past outcomes.
* Techniques: Drill-down, correlation analysis.
* Example: Analyzing why sales dropped in a region.
* Business Value: Helps understand underlying problems.
3.
Predictive Analytics:
* What is likely to happen?
* Uses statistical models and ML to forecast future events.
* Example: Predicting customer churn.
* Business Value: Enables proactive decision-making.
4.
Prescriptive Analytics:
* What should be done?
* Recommends actions using optimization and simulation.
* Example: Suggesting the best combination of products for
cross-selling.
* Business Value: Guides strategic actions for maximum
impact.
Comparison:
| Type |
Timeframe | Key Question |
Example |
| ------------ | --------- | -------------------- |
------------------------- |
| Descriptive |
Past | What happened? | Revenue report |
| Diagnostic |
Past | Why did it happen? | Regional performance drop |
| Predictive |
Future | What will happen? | Sales forecast |
| Prescriptive | Future
| What should be done? | Price optimization |
Q8. Discuss the role of predictive analytics in customer
relationship management. Illustrate with relevant case studies.
Answer:
Predictive analytics enhances Customer Relationship
Management (CRM) by forecasting customer behavior and enabling personalized
interactions.
Applications:
* Customer Segmentation: Groups customers by behavior or
demographics.
* Churn Prediction: Identifies customers likely to leave.
* Personalized Recommendations: Suggests products tailored
to preferences.
* Customer Lifetime Value (CLV): Estimates long-term value
of a customer.
* Upselling/Cross-selling: Identifies opportunities to sell
related products.
Case Study 1: Amazon
Amazon uses predictive models to recommend products based on
browsing and purchase history, increasing engagement and repeat purchases.
Case Study 2: Salesforce Einstein
Einstein AI in Salesforce predicts which leads are more
likely to convert and recommends actions to improve success.
Case Study 3: Vodafone
Implemented a churn prediction model that accurately
identified high-risk customers and offered retention deals, reducing churn by
20%.
Q9. Evaluate the current state of predictive analytics
adoption across industries. What factors contribute to successful
implementation?
Answer:
Current State:
* Widespread adoption across industries such as retail,
banking, healthcare, telecom, and logistics.
* Growth in investment in AI and analytics platforms.
* Organizations moving from descriptive to predictive and
prescriptive models.
Success Factors:
1. Clear Business Objective: Predictive models aligned with
strategic goals perform better.
2. High-Quality Data: Reliable, clean, and integrated data
is essential.
3. Skilled Workforce: Data scientists, analysts, and domain
experts are key.
4. Organizational Culture: Support for data-driven
decision-making from leadership.
5. Scalable Infrastructure: Cloud computing and big data
tools support real-time analytics.
6. Robust Governance: Ensures data security, ethics, and
compliance.
Example: Netflix's success with personalized recommendations
and content investment decisions is due to their data-first culture and
analytics capabilities.
Q10. Analyze how predictive analytics can provide
competitive advantage in today's data-driven market environment with
appropriate examples.
Answer:
Predictive analytics offers a competitive edge by enabling
organizations to:
* Anticipate customer needs and improve personalization.
* Reduce costs through optimized operations.
* Detect and prevent risks like fraud or churn.
* Innovate by identifying new trends early.
Examples:
1. Netflix: Uses predictive models for content
recommendation and greenlighting new shows, increasing viewer satisfaction and
retention.
2. Uber: Predictive analytics informs surge pricing, driver
allocation, and estimated time of arrival.
3. Walmart: Forecasts demand to optimize inventory, reducing
waste and increasing shelf availability.
4. American Express: Predicts customer churn and offers
timely retention offers.
By turning data into foresight, companies outperform those
relying only on intuition or historical reporting.
—
Next: I'll continue with Questions 11–15 in the following
message.
Here are the full answers for Questions 11 to 20 from the
CSE3036 – Predictive Analytics question bank.
Q11. Explain the concept of analytics, its importance in
today's business environment, and discuss how it aids in strategic
decision-making with relevant examples.
Answer:
This question is a repetition of Q1. As already explained,
analytics helps in transforming data into actionable insights through
descriptive, predictive, and prescriptive techniques. Its importance lies in
improving operational efficiency, enhancing customer satisfaction, and
fostering innovation.
Example: Domino's Pizza uses analytics to forecast order
demand and staff scheduling, improving delivery speed and reducing food waste.
Q12. Demonstrate the evolution of analytics as a field and
explain why it has gained such popularity among businesses in the past decade.
Answer:
Analytics has evolved through four stages:
* Descriptive Analytics: Summarizes historical data.
* Diagnostic Analytics: Explores reasons behind outcomes.
* Predictive Analytics: Forecasts future trends.
* Prescriptive Analytics: Recommends optimal actions.
In recent years, analytics has surged in popularity due to:
* Availability of big data.
* Advancements in computing power.
* Machine learning breakthroughs.
* Demand for competitive advantage.
Example: Spotify's use of analytics to recommend music and
personalize playlists has transformed user engagement.
Q13. Explain the concept of propensity models in detail,
discussing their types, applications, and methodology of development with
appropriate examples.
Answer:
Propensity models estimate the likelihood of a customer
performing a certain action, such as buying, leaving, or subscribing.
Types:
* Propensity to Buy: Who is likely to purchase.
* Propensity to Churn: Who is likely to leave.
* Propensity to Upsell: Who is likely to buy more.
Development Methodology:
1. Collect customer data.
2. Select target behavior.
3. Preprocess data and engineer features.
4. Train a classification model (e.g., logistic regression).
5. Validate and interpret scores.
Example: An insurance company uses a propensity model to
predict which customers are likely to renew their policies and targets
high-risk individuals with incentives.
Q14. Analyze the working principles of collaborative
filtering systems. Compare and contrast user-based and item-based collaborative
filtering approaches with examples.
Answer:
Collaborative filtering (CF) recommends products based on
similarities in user behavior or item consumption.
1.
User-based CF:
* Recommends items liked by similar users.
* Example: If User A and User B have similar preferences,
recommend what B liked to A.
2.
Item-based CF:
* Recommends items similar to those the user liked.
* Example: If A liked Item X and Item Y is similar,
recommend Y.
Comparison:
| Aspect |
User-Based CF |
Item-Based CF |
| ----------- | ----------------------------------- |
------------------------------------- |
| Scalability | Lower | Higher |
| Stability | Less
stable (user behavior changes) | More stable (items don't change much) |
| Example |
MovieLens system |
Amazon "also bought" |
Both techniques suffer from the cold start problem for new
users or items.
Q15. Discuss the complete process of cluster modeling,
including different algorithms, evaluation methods, and business applications.
Illustrate with a case study.
Answer:
Clustering is an unsupervised learning technique that groups
similar data points.
Process:
1. Data Preprocessing: Normalize and clean data.
2. Select Algorithm:
* K-Means:
Partitions data into k clusters.
* Hierarchical:
Builds nested clusters.
* DBSCAN:
Identifies arbitrary-shaped clusters.
3. Evaluate:
* Silhouette Score.
* Davies-Bouldin
Index.
4. Interpret Results:
·
Assign business meaning to each cluster.
Applications:
* Customer segmentation.
* Fraud detection.
* Market basket analysis.
Case Study: A telecom company uses K-means to segment users
into budget users, premium users, and business users, enabling personalized
plans and reducing churn.
Q16. Explain univariate and multivariate statistical
analysis techniques. Compare their methodologies, applications, and limitations
in the context of predictive analytics.
Answer:
Univariate Analysis:
* Involves one variable.
* Describes distribution, central tendency, dispersion.
* Example: Analyzing average income.
Multivariate Analysis:
* Involves multiple variables.
* Explores relationships (e.g., correlation, regression).
* Example: Predicting income based on age, education, and
experience.
Comparison:
| Aspect |
Univariate |
Multivariate
| ---------- | ----------------------- |
----------------------------------------------- |
| Variables |
One | Two or
more
| Output |
Descriptive |
Predictive/Explanatory
| Techniques | Histograms, mean, SD | Regression, factor analysis, MANOVA |
| Limitation | No interaction insights | Assumes
independence, risk of multicollinearity |
In predictive analytics, multivariate models provide deeper
insights but require careful validation.
Q17. Critically evaluate the limitations of predictive
modeling and discuss strategies to overcome these limitations in practical
business scenarios.
Answer:
Limitations:
* Overfitting: Model performs well on training data but
poorly on new data.
* Underfitting: Model too simple to capture data patterns.
* Data Quality: Garbage in, garbage out.
* Model Bias: May reflect historical discrimination.
* Lack of Explainability: Complex models like neural
networks are black boxes.
Strategies:
* Cross-validation to detect overfitting.
* Regularization (Lasso, Ridge) to simplify models.
* Clean and enrich data sources.
* Use explainable AI tools like SHAP and LIME.
* Monitor and update models regularly.
Example: A retail chain retrains its demand forecasting
model quarterly to adjust for seasonal shifts and market dynamics.
Q18. Describe the statistical foundations necessary for
effective predictive analytics. Include discussions on probability,
distributions, and hypothesis testing.
Answer:
1.
Probability:
* Foundation for classification models (e.g., Naïve Bayes).
* Concepts include conditional, joint, and marginal
probabilities.
2.
Statistical Distributions:
* Normal distribution: Common in regression and control
charts.
* Poisson: Used in queuing and event count modeling.
* Binomial: Applicable to binary outcomes.
3.
Hypothesis Testing:
* Tests claims using data evidence.
* Common tests: t-test, chi-square, ANOVA.
* Example: A/B testing for campaign effectiveness.
These tools support model selection, validation, and
understanding of relationships between variables.
Q19. Analyze how predictive modeling techniques are applied
in customer segmentation. Discuss the methodology, variable selection, and
business implications with examples.
Answer:
Methodology:
1. Data Collection: Demographics, behavior, transactions.
2. Feature Selection: RFM (Recency, Frequency, Monetary),
geography, product use.
3. Choose Model: K-means, DBSCAN for clustering.
4. Evaluate and Label Clusters.
5. Use segments for targeting.
Business Implications:
* Personalized marketing.
* Improved resource allocation.
* Better customer retention.
Example: An online retailer segments customers into VIPs,
discount-seekers, and window shoppers. Tailored emails to each group increase
conversion rates.
Q20. Explain the mathematical principles behind various
statistical techniques used in predictive analytics and their practical
implementations.
Answer:
1.
Linear Regression:
* Models relationship as: Y = β₀ + β₁X + ε
* Minimizes sum of squared residuals.
2.
Logistic Regression:
·
Models probability: P(Y=1) = 1 / (1 + e^-(β₀ +
β₁X))
3.
Decision Trees:
·
Splits data using metrics like entropy or Gini
index.
4.
K-Means:
·
Minimizes within-cluster sum of squares.
5.
PCA (Principal Component Analysis):
·
Reduces dimensionality by transforming variables
into orthogonal components.
Practical Use:
* Regression in forecasting.
* Logistic models for classification.
* Decision trees in credit scoring.
These techniques are implemented in Python libraries
(scikit-learn, statsmodels) and used across industries for predictive tasks.
—
Next: I will continue with full answers for Questions 21–30
in the next message.
Here are the full answers for Questions 21 to 35 from the
CSE3036 – Predictive Analytics question bank:
Q21. Discuss the importance of feature engineering and
variable selection in multivariate analysis. How do these processes impact
model performance?
Answer:
Feature engineering transforms raw data into meaningful
input variables for modeling. It includes:
* Encoding categorical variables.
* Creating interaction features (e.g., age × income).
* Handling missing values.
* Scaling and normalizing features.
Variable selection helps reduce dimensionality and improves
model accuracy and interpretability.
Techniques:
* Filter: Correlation, mutual information.
* Wrapper: Recursive feature elimination.
* Embedded: LASSO, Ridge.
Impact:
* Reduces overfitting.
* Speeds up training.
* Enhances interpretability.
* Improves predictive performance.
Example: In a loan default model, "debt-to-income ratio" may
be a more informative feature than income alone.
Q22. Critically evaluate the limitations of predictive
modeling and discuss strategies to overcome these limitations in practical
business scenarios.
Answer:
(Already covered in Q17. Refer back for detailed
explanation.)
Q23. Explain the mathematical principles behind various
statistical techniques used in predictive analytics and their practical
implementations.
Answer:
(Already covered in Q20. See above.)
Q24. Critically evaluate linear regression and its variants
(ridge, lasso, elastic net). Discuss their mathematical foundations and
applications with examples.
Answer:
Linear Regression:
Y = β₀ + β₁X₁ + … + βnXn + ε. It minimizes squared errors.
Assumes linearity and homoscedasticity.
Ridge Regression (L2):
Penalizes sum of squares of coefficients: λΣβ². Shrinks
coefficients to prevent overfitting.
Lasso Regression (L1):
Penalizes sum of absolute coefficients: λΣ|β|. Performs
feature selection.
Elastic Net:
Combination of L1 and L2. Useful when predictors are
correlated.
Applications:
* Ridge: Multicollinear data.
* Lasso: Sparse feature sets.
* Elastic Net: High-dimensional data (e.g., genomics).
Q25. Demonstrate a comprehensive analysis of non-linear
regression models including polynomial regression.
Answer:
Non-linear models capture curved relationships.
Polynomial Regression:
Y = β₀ + β₁X + β₂X² + … + βnXⁿ + ε.
* Allows modeling of U-shaped curves.
* Risk: Overfitting with high-degree polynomials.
Other Non-linear Forms:
* Exponential: Y = ae^(bX)
* Logarithmic: Y = a + b ln(X)
* Power: Y = aX^b
Applications:
* Polynomial: Pricing models.
* Exponential: Growth modeling.
* Logarithmic: Learning curves.
Q26. Discuss classification performance metrics beyond
accuracy. Explain their significance in imbalanced datasets with appropriate
examples.
Answer:
Accuracy is misleading when class distribution is skewed.
Better metrics:
* Precision = TP / (TP + FP): Low FP.
* Recall = TP / (TP + FN): Low FN.
* F1 Score = 2PR / (P + R): Balance of precision and recall.
* AUC-ROC: Measures classifier's ability to rank
predictions.
Example: In spam detection, high recall ensures most spam is
caught, while high precision avoids misclassifying legitimate emails.
Q27. Compare and contrast different classification
algorithms including logistic regression and decision trees. Analyze their
strengths and weaknesses.
Answer:
Logistic Regression:
* Linear decision boundary.
* Simple, interpretable.
* Poor with complex data.
Decision Trees:
* Non-linear splits.
* Easy to visualize.
* Overfits on small data.
Others:
* SVM: Great in high dimensions.
* k-NN: Simple, needs scaling.
* Naïve Bayes: Good with text; assumes feature independence.
Comparison Table:
| Algorithm |
Interpretability | Performance |
Complexity |
| ------------- | ---------------- | -------------------- |
---------- |
| Logistic |
High | Medium | Low |
| Decision Tree | High | High risk of overfit |
Medium |
| SVM |
Low | High | High |
Q28. Compare and contrast supervised and unsupervised
learning methods in detail, providing examples of algorithms from each category
and their appropriate applications.
Answer:
| Type |
Supervised |
Unsupervised
| Input |
Features + Labels | Features
only |
| Goal | Predict
target | Discover patterns |
| Algorithms | Regression, SVM, Trees | K-Means, PCA, DBSCAN |
| Use Cases | Churn
prediction, sales | Customer segmentation, anomaly detection |
Examples:
* Supervised: Predicting if a customer will default.
* Unsupervised: Clustering users into marketing segments.
Q29. Discuss the importance of cross-validation in model
selection. Explain different cross-validation techniques and how they help in
building robust predictive models.
Answer:
Cross-validation evaluates model generalizability.
Techniques:
* Holdout: Simple split.
* k-Fold: Train on k-1 folds, test on 1; repeat.
* Stratified k-Fold: Maintains class distribution.
* LOOCV: Leave-one-out; high variance.
* Time Series CV: Maintains temporal order.
Importance:
* Reduces overfitting.
* Supports hyperparameter tuning.
* Provides reliable performance estimates.
Q30. Analyze the bias-variance trade-off in predictive
modeling. How does this concept influence model complexity and what strategies
can be employed to find the optimal balance?
Answer:
Bias: Error from overly simplistic assumptions. High bias
underfits.
Variance: Error from model sensitivity to data fluctuations.
High variance overfits.
Trade-off:
* Simple model: High bias, low variance.
* Complex model: Low bias, high variance.
Strategies:
* Use cross-validation to monitor performance.
* Apply regularization (L1/L2).
* Prune decision trees.
* Use ensemble methods (bagging reduces variance, boosting
reduces bias).
Q31. Explain the methodology of selecting the most
appropriate model for a given business problem. Include discussions on business
constraints, data characteristics, and model complexity.
Answer:
Methodology:
1. Understand business objective.
2. Analyze data type, size, structure.
3. Test baseline models (e.g., logistic regression, decision
trees).
4. Evaluate with metrics aligned to goal (e.g., F1, AUC,
RMSE).
5. Consider constraints:
* Interpretability.
* Inference speed.
* Data
availability.
Example: For a healthcare diagnostic tool, logistic
regression may be preferred over deep learning for explainability.
Q32. Discuss techniques for handling imbalanced datasets in
classification problems.
Answer:
Techniques:
·
Resampling:
* Oversample
minority (SMOTE).
* Undersample
majority.
* Algorithmic:
* Adjust class
weights.
* Use cost-sensitive
learning.
* Ensemble:
* Balanced Random
Forest.
* Evaluation:
·
Use precision, recall, F1 instead of accuracy.
Example: In fraud detection (1% fraud), SMOTE increases
minority class representation, improving recall.
Q33. Analyze techniques for handling non-stationary time
series data. Include discussions on transformations, differencing, and testing
methods with examples.
Answer:
Non-stationary data has changing mean/variance.
Techniques:
* Differencing (ΔYₜ = Yₜ - Yₜ₋₁).
* Log or Box-Cox transformations for variance.
* Detrending via regression.
* Seasonal Differencing (ΔYₜ = Yₜ - Yₜ₋ₛ).
Testing:
* Augmented Dickey-Fuller (ADF): H₀ = non-stationary.
* KPSS: H₀ = stationary.
Example: Air passenger data requires log + seasonal
differencing to stabilize before ARIMA modeling.
Q34. Evaluate the application of time series analysis in
demand forecasting. Discuss methodology, variable selection, and evaluation of
forecast accuracy with a business case study.
Answer:
Methodology:
1. Data prep: Remove outliers, fill missing.
2. Model: ARIMA, Prophet, LSTM.
3. Feature selection: Lagged values, seasonality, holidays.
4. Evaluate: MAE, RMSE, MAPE.
Case Study: Walmart uses ARIMA + promotions/holiday
variables to forecast weekly store sales, reducing overstock and improving
customer satisfaction.
Q35. Explain the various performance metrics used for
evaluating regression models. Discuss their calculations, interpretations, and
situations where each would be most appropriate.
Answer:
* MAE = mean(|yᵢ - ŷᵢ|): Intuitive; robust to outliers.
* MSE = mean((yᵢ - ŷᵢ)²): Penalizes large errors.
* RMSE = √MSE: Same units as output; sensitive to outliers.
* R² = 1 – SSR/SST: Proportion of variance explained.
* Adjusted R²: Adjusts for number of predictors.
Use MAE when outliers are not critical, RMSE when large
errors matter, and R² for overall model fit.
—
Next, I'll continue with the final set of answers: Questions
36–47. Let me know if you'd like that now.