Solved Paper - Data Mining & Warehousing

Q1(a) Framework of Data Warehouse:

A typical data warehouse has four components: Operational Database, ETL Process, Data

Warehouse (staging, integration, access layers), and front-end tools. [Diagram is usually required].

Q1(b) Dimensional Model:

A data structure optimized for data warehousing tools. Design steps: Choose business process,

Declare grain, Identify dimensions, Identify facts.

Q1(c) OLTP vs OLAP:

OLTP: Real-time, high volume, normalized data.

OLAP: Analytical, historical data, denormalized schema.

Q1(d) EDW Parts:

1. Data Sources, 2. ETL Tools, 3. Staging Area, 4. Data Storage, 5. Metadata, 6. Query Tools.

Q1(e) Star Schema Example:

Fact Table: Transactions (Amount, Date, AccountID)

Dimension Tables: Customer, Time, Branch, AccountType

Q1(f) Hybrid DW Model:

Used when combining top-down and bottom-up approaches. Preferred when flexibility and faster

implementation are required.

Q2(a) Data to be mined:

Patterns, associations, clusters, outliers, predictive models.

Support, Confidence, Lift, Accuracy, Precision, Recall, F-measure.

Q2(c) Statistical Description:

Includes measures of central tendency (mean, median), dispersion (variance, std deviation), and

distribution.

Q2(d) Need for Data Cleaning:

To remove noise, handle missing values, correct inconsistencies and improve data quality.

Q2(e) Apriori Algorithm:

Frequent itemsets: {3}, {5}, {2,5}, {1,3}

Rules example: {2}->{5}, Support=50%, Confidence=75%

Q2(f) FP-Growth Tree:

1. Count frequency.

2. Order items.

3. Build tree level-wise.

4. Extract patterns from tree.

Q3(a) Classification vs Prediction:

Classification predicts categorical labels, prediction forecasts continuous values.

Q3(b) Linear Regression:

Models relationship as Y = aX + b. E.g., Predicting sales based on advertising spend.

Metrics: Accuracy, Confusion Matrix, ROC Curve, Precision, Recall.

Q3(d) K-means Steps:

1. Choose k

2. Assign points

3. Update centroids

4. Repeat till convergence.

Q3(e) ID3 Algorithm:

Build decision tree using information gain. Root: Age. Classification: Uses best attributes to classify

buys_computer.

Q3(f) Clustering Applications:

Marketing, Insurance Fraud Detection, Document Categorization, Customer Segmentation.

Q4(a) Features of Data Warehouse:

Subject-oriented, Integrated, Time-variant, Non-volatile.

Q4(b) Attribute Types:

Nominal, Ordinal, Interval, Ratio.

Q4(c) Clustering Requirements:

Scalability, Ability to deal with noise, Interpretability, High dimensionality support.

Q4(d) Granularity of Facts:

Level of detail. Fine granularity gives detailed data. Coarse granularity is summarized.
Q4(e) Association Rule Metrics:

Support: Frequency of itemset. Confidence: Likelihood of consequent given antecedent. Risk: Often

linked with lift or leverage.

Q4(f) Classification Applications:

Spam Detection, Medical Diagnosis, Customer Churn, Credit Scoring.