SVM

The idea:

What Logistic regression is draw a line between two classes. SVM is like that line but with a margin. SVM tries to find the 'maximum marginal classifier'

The math is non-trivial : Convex optimization (Stanford has a course on this) SVM fit takes a long time, but once it's done it's reasonably quick.

Neural network usually outperforms SVM.

The model has to store the support vectors, or the vector of values at the margin.

kernel = 'linear' logistic 'poly' circle 'rbf' radio basis functions (use as default (the best) - local probability . local gaussian with a margin.)

SVM Here's your weight - c All your weights when you sum up should be less than c

Giving it some budget(wiggle room) for it to go outside a margin

There's SVR and SVC