Question 1

What is the 10-step Data Science Methodology developed by John B. Rollins?

Accepted Answer

John B. Rollins (Data Scientist at IBM Analytics) developed a 10-step iterative framework grouped into five two-stage modules: 1) From Problem to Approach (Business Understanding & Analytic Approach); 2) From Requirements to Collection (Data Requirements & Data Collection); 3) From Understanding to Preparation (Data Understanding & Data Preparation); 4) From Modelling to Evaluation (AI Modelling & Evaluation); and 5) From Deployment to Feedback (Deployment & Feedback).

Question 2

What are the four types of Data Analytics in Unit 2?

Accepted Answer

The four types are: 1) Descriptive Analytics (What happened? Summarizes historical data using statistics and charts); 2) Diagnostic Analytics (Why did it happen? Root cause analysis and hypothesis testing); 3) Predictive Analytics (What will happen next? Regression and classification forecasting); and 4) Prescriptive Analytics (How do we solve it / what action to take? Optimization and decision simulation).

Question 3

What is the difference between Train-Test Split and K-Fold Cross-Validation?

Accepted Answer

Train-Test Split divides data into two distinct sets (e.g. 80% train, 20% test) and is best suited for large datasets with clear demarcation. K-Fold Cross-Validation divides data into k subsets (folds), iteratively training on k-1 folds and validating on the remaining fold until every data point has been in both training and test sets. Cross-validation gives a more reliable estimate of model generalizability on small datasets.

Question 4

How are Classification metrics calculated from a Confusion Matrix?

Accepted Answer

From a 2x2 confusion matrix with TP, TN, FP, and FN: Precision = TP / (TP + FP); Recall = TP / (TP + FN); F1-Score = 2 * (Precision * Recall) / (Precision + Recall); and Accuracy = (TP + TN) / (TP + FP + FN + TN).

Question 5

What are the core evaluation metrics for Regression models?

Accepted Answer

Regression metrics evaluate numeric errors between actual (y) and predicted (y-hat) values: 1) Mean Absolute Error (MAE); 2) Mean Squared Error (MSE), which squares errors penalizing large deviations heavily; and 3) Root Mean Squared Error (RMSE), the square root of MSE, preferred because it is expressed in the exact same units as the target variable.

Data Science MethodologyAn Analytic Approach to Capstone Project

2.1 The Foundational Data Science Methodology (John B. Rollins)

From Problem to Approach

Requirements to Collection

Understanding to Preparation

Modelling to Evaluation

Deployment to Feedback

Next: Unit 3 · Making Machines See