The DL2 (Euclidean distance) is known to be less robust against outliers than the DL1 (city
block distance) [3]. We have chosen to use DRL1 which weights the contribution of each
element to DL1 according to the magnitudes.
Classification Performance: Let N
w1
t
, N
w2
t
show the number of test vectors in set Tt which
are known to belong to the w1 and w2 classes, respectively. For a diagnosis problem like
this, the analysis of four ratios can represent our method’s performance. Let t pwi
, f nwi
show the number of correct and missed classifications for class wi respectively.