This repository contains well-structured Viva Questions and Answers for all the lab assignments from the Data Mining and Predictive Modeling (DMPM) course.
These Q&As will help you revise concepts and confidently face your practical viva.
- Data Exploration & Visualization
- Linear Regression
- Logistic Regression
- Decision Tree
- K-Means Clustering
- Association Rule Mining
Q: What is the purpose of data exploration?
A: To understand the dataset, identify patterns, and detect missing values or outliers.
Q: How did you handle missing values?
A: I used mean or median imputation for numerical columns.
Q: What is skewness and kurtosis?
A: Skewness indicates asymmetry; kurtosis indicates the heaviness of tails in the distribution.
Q: Why log-transform numerical features?
A: To normalize skewed data and make it suitable for statistical analysis.
Q: How did you split the data?
A: Using train_test_split with a 70-15-15 ratio for training, validation, and testing.
Q: What does linear regression do?
A: Predicts a continuous dependent variable based on independent variables.
Q: How did you evaluate your model?
A: Using Mean Squared Error (MSE) and R² Score.
Q: What are residuals?
A: The difference between actual and predicted values. Useful for model diagnostics.
Q: What does an R² Score indicate?
A: The proportion of variance in the target explained by the model.
Q: When do you use logistic regression?
A: For binary classification problems.
Q: What metrics did you use to evaluate the model?
A: Accuracy, Confusion Matrix, Precision, Recall, F1-score, ROC AUC Score.
Q: What does the ROC curve represent?
A: The trade-off between true positive rate and false positive rate at various thresholds.
Q: How did you define the target variable?
A: mpg_class = 1 if mpg > median, else 0.
Q: Why use a decision tree?
A: It's easy to interpret, doesn’t need feature scaling, and works well for both classification and regression.
Q: How did you validate the model?
A: With accuracy metrics and 5-fold cross-validation.
Q: What is GridSearchCV?
A: A method to find the best hyperparameters using cross-validation.
Q: What is the goal of K-Means?
A: To group similar data points into k clusters based on distance metrics.
Q: What is inertia in the Elbow Method?
A: The sum of squared distances of samples to their closest cluster center.
Q: Why use PCA in clustering?
A: To reduce dimensions for better visualization.
Q: What is an association rule?
A: A rule that implies a relationship between items in a transaction dataset (e.g., {Bread} → {Butter}).
Q: What do support, confidence, and lift mean?
- Support: Frequency of itemset in dataset
- Confidence: Probability of consequent given antecedent
- Lift: Strength of rule over random occurrence
Q: How did you implement it?
A: Using mlxtend.apriori() for frequent itemsets and association_rules() for rules.
Q: Why sort by lift?
A: To prioritize the most meaningful and non-random rules.
Made with ❤️ by VED
For academic purposes under the course Data Mining & Predictive Modeling
- Python
- Pandas, NumPy
- Scikit-learn
- Seaborn, Matplotlib
- mlxtend