5. k-Nearest Neighbor Classification (Geometric and Probabilistic Interpretation).pdf

k-Nearest Neighbor Classification ► potential problem: nearest neighbor might be an outlier, somehow unusual, misleading 0 0° 0...

k-Nearest Neighbor Classification ► potential problem: nearest neighbor might be an outlier, somehow unusual, misleading 0 0° 0 ► take more than one neighbor into consideration: k nearest neighbor classifier decision set: the set of (k) nearest neighbors considered for the classification decision decision rule: how to decide the class, given the potentially different classes of the k nearest neighbors ► take the majority vote ► potentially weighted votes Let x1,... ,xk be the k nearest neighbors of instance xq: k 1 if a= b h(xq ) = argmax L cEC. i=l wi8(c,f(xi)) where 8(a,b) = { O otherwise