the interpretability and assesment of the results

we need a human expert to label the data

the datasets are expensive to collect

it’s impossible to gather dataset big enough to represent real world

beyond a certain dimensionality of the feature space the model starts to deteriorate instead of improving

it’s impossible to select optimal dimension of feature space

OR

NOT

XOR

AND

we don’t have any model of the environment and only way to collect information is to interact with it

we have only a simulation model of the environment

environment is well modelled, but we don’t have analytic solution

methods that work with incomplete and imprecise data, giving highly probable results

methods that give accurate and precise results, but are computationally very demanding

methods that save computational power, but give imprecise and inaccurate results

can be used for both classification and regression

is immune to noise in feature values

is a method where the result is based on k closest training examples

is a method of supervised learning

proper choice of the metric

normalization of features

using weighted voting

feature extraction