lecture 10 imbalanced data

Question 1

random undersampling

Answer 1

Drop data
Fast training
Loses data

Question 2

random oversampling

Answer 2

Repeat sample
Much slower

Question 3

class weight

Answer 3

Reweight the loss function
Same effect as oversampling , not as expensive

Question 4

ensemble resampling

Answer 4

Random resampling separate for each instance in an ensemble

Question 5

edited nearest neighbors

Answer 5

Reducing dataset for knn
Remove Al samples that are mis classified by knn from training
Cleans up outliers and boundaries

Question 6

condensed nearest neighbors

Answer 6

Add linings to the data that are mis classified by knn
Focus on the boundaries
Removes many

Question 7

synthetic sample generator

Answer 7

Add synthetic interpolated data to smaller class
For each sample in minority class
Pick random neighbor from k neighbors
Pick point on line connecting the two uniformly
Large dataset
Combined with under sampling strategies