predictive analytics; prediction Flashcards
1
Q
confidence interval
A
quantifies uncertainty surrounding mean over a large quantity
2
Q
prediction interval
A
quantifies uncertainty surrounding prediction of single quantity
3
Q
machine learning definition
A
study of algorithms applied to data, focussing on prediction and classification
4
Q
machine learning characteristics
A
- formulae like regression, but rules to produce yhat
- translated into computer code and automated
- predictions generated quickly and repeated without intervention
- takes error out of humans choosing variables
- challenges of interpretability and bias
5
Q
regression process
A
- hypothesis
- select variables
- train
- test
- select best model
- refine
6
Q
problems of machine learning
A
1) good for prediction, not explanatory
2) black box (don’t know why machine chooses variables)
3) ethical (sub-optimal outcomes like discrimination)
7
Q
black swan
A
extreme outliers that have a disproportional effect (overfitting)
8
Q
overfitting
A
- only see particular subset of data, and what is true is smth bigger we don’t observe
- trying so hard to explain what we see, so poor at explaining what we don’t see
9
Q
avoiding overfitting
A
- split data into training and testing
- training: subset of data, eg 80% to estimate formula
- testing: 20% of data to test how well model predicts variables not shown
10
Q
R squared
A
increases with number of variables
11
Q
adjusted R squared
A
- penalises for number of variables
- asks if I gain much explanatory power for extra variables
- can go down if irrelevant variable is included
- bigger is better = more accurate
12
Q
mean squared error
A
- common mean of selecting model
- smaller is better = less error
- best model = one with least errors
- high 𝑅^2 implies low MSE
13
Q
regression tree
A
- output from machine learning
- segments inputs in mutually exclusive/exhaustive regions
- branches connect nodes to a terminal node (leaf)
