Other classical approaches.
WHAT NEAREST NEIGHBOUR METHODS DO
Predict from the most similar training examples.
WHAT THEY PROVIDE
No training, and natural handling of complex boundaries.
WHAT THEY COST
Slow prediction, growing with dataset size Sensitivity to scaling and irrelevant features
WHERE THEY REMAIN USEFUL
Recommendation, similarity search, and small datasets.
WHAT SUPPORT VECTOR MACHINES DO
Find the boundary maximising separation between classes.
WHAT KERNELS PROVIDE
Handling non-linear boundaries without explicit transformation.
WHERE THEY SUIT
Small to medium datasets with clear margins.
WHAT THEY COST
Poor scaling to very large datasets.
WHAT NAIVE BAYES DOES
Applies probability with an assumption that features are independent.
WHY IT WORKS DESPITE THAT ASSUMPTION BEING FALSE
Classification only requires the right class to score highest, not calibrated probabilities.
WHERE IT SUITS
Text classification, and situations requiring extreme speed.
WHAT CLUSTERING METHODS DIFFER IN
Whether the number of clusters is specified Whether clusters must be round Whether every point must belong to one
WHAT TO CHOOSE ON
The shape of the structure you expect.
