1/29
BA 260 Final Exam
Supervised Learning
Uses known outcomes to predict outcomes for new cases.
Unsupervised Learning
Finds patterns or groups without a target outcome.
Data Preparation
Cleans, transforms, and organizes data before modeling.
Scatter Chart
Plots two variables to show their relationship.
Training Data
Data used to build the model.
Validation Data
Data used to tune or compare models.
Test Data
Held-back data used for a final performance check.
Overfitting
When a model fits training data well, but performs poorly on new data.
True Positive
Predicted positive; actually positive.
False Positive
Predicted Positive; actually negative.
True Negative
Predicted negative; actually negative.
False Negative
Predicted negative; actually postitive.
Sensitivity
Fraction of actual positives correctly identified:
TP/(TP+FN)
(recall)
Precision
Fraction of predicted positives that are truly positive:
TP/(TP+FP)
Accuracy
Fraction of predictions that are correct;
1 - error rate
Error Rate
Fraction of predictions that are incorrect.
Cuttoff Value
Probability threshold that turns a model’s score into a positive or negative prediction.
Lower Cutoff
Usually flags more positives, raising sensitivity and false positives.
Higher Cutoff
Usually flags fewer positives, reducing false positives, but potentially missing more actual positives.
Data Exploration
Uses summaries and visuals to understand data and patterns.
Supervised
A bank uses past loan records labeled “repaid” or “defaulted” to predict whether new applicants will default.
What learning is this?
Unsupervised
A store groups shoppers by similar shopping habits, without predicting a labeled outcome.
What learning is this?
Data Preparation
Correcting typos and handling missing values before modeling.
Data Exploration
Creating charts and summary statistics to understand the data.
TP/(TP+FN)
What is the formula for sensitivity?
TP/(TP+FP)
What is the formula for precision?
85%
A model’s error rate is 15%.
What is its accuracy?
More, Increases
If you lower the classification cutoff, does the model usually flag more cases or fewer cases as positive?
What usually happens to false positives?
Lower, Rises
A hospital wants to catch more patients who truly have a disease and can accept extra false alarms.
Should it raise or lower the cutoff?
What happens to sensitivity?