End-of-lesson Quiz
5 questions · Linear Regression
1
of 5
Linear regression is best suited for predicting which kind of value?
Linear regression outputs a real number with no upper bound. Categories are the job of logistic regression or trees; bounded probabilities are logistic; clusters come from unsupervised methods like K-means.
2
of 5
In
y = w·x + b, what are w and b after training?
For one-feature linear regression, the model is those two numbers. Once fit, you can throw the training data away and ship just
w and b. That's the whole point of step 1: a model is two numbers.
3
of 5
Why do we split data into a train set and a test set?
The test set simulates data the model has never seen. If train R² is much higher than test R², the model has memorised the training rows and will fail in production — classic overfitting.
4
of 5
Your response-time model reports RMSE = 15 ms. What does that mean?
RMSE is in the same units as the target. ~15 ms is the typical magnitude of prediction error — useful because you can compare it directly to the response times you're predicting.
5
of 5
You're using
kσ residual thresholding to flag slow responses. What does the choice of k trade off?
Lower
k (e.g. 2σ) catches more anomalies but also fires on more benign noise. Higher k (e.g. 3σ) is conservative but misses subtle attacks. This is the classic alert fatigue tradeoff every SOC faces.