SDS Data science techniques and modelling

A model that fits the training data very closely but performs poorly on new data is:

• 1 Unbiased
• 2 Well calibrated
• 3 Underfitted
• 4 Overfitted

A fraud model flags 50 of 1,000 transactions. 30 of the flagged are real fraud, and there are 40 frauds in total. The precision is:

• 1 97%
• 2 75%
• 3 30%
• 4 60%

A fraud model flags 50 of 1,000 transactions. 30 of the flagged are real fraud, and there are 40 frauds in total. The recall is:

• 1 75%
• 2 60%
• 3 97%
• 4 50%

Accuracy can be a poor measure of performance when:

• 1 The model is simple
• 2 There are many columns
• 3 Data is numeric
• 4 One class is very rare

Grouping customers into segments without predefined labels is an example of:

• 1 Unsupervised learning (clustering)
• 2 Regression
• 3 Supervised classification
• 4 Hypothesis testing

Predicting a customer's monthly spend as a number is a:

• 1 Anomaly detection problem
• 2 Classification problem
• 3 Regression problem
• 4 Clustering problem

Splitting data into training and test sets is done to:

• 1 Reduce the file size
• 2 Avoid cleaning
• 3 Hide the data
• 4 Estimate how the model will perform on unseen data