M. Járó - L. Költő szerk.: Archaeometrical research in Hungary (Budapest, 1988)
Analysis - GEGUS Ernő, BORSZÉKI János: Investigation of archaeological metal findings by a laser-microspectral analysis method and characterization of results using pattern recognition methods
After selecting the most important features, the classification of samples or the ranging into classes can be performed in the field of significant features. Hierarchic classification is a method demanding considerable computer capacity but often supplying good results. With this technique the distances in space between the sample points are determined by using the Euclidean distance function: ELXx,y)= 2 (xi-yO 2 i=l 1/2 ED (x,y) the Euclidean distance of sample points x and y in the n-dimensional space. Using the Euclidean distances of sample points a dendrogram will be plotted on which the formation of groups can be established on the basis of similarity of the sample points. As a similarity function, the following equation is generally used: 1J ED max where Sy is the similarity value of the/'th and/th sample points, EDjj the Euclidean distance of /th and /'th sample points, ED m ax the maximum Euclidean distance. With the task of classifying a sample of unknown origin or considered unknown in whatever respect, the K-nearest-neighbours (KNN-) method may be used.The classification of an unknown sample into classes of known samples is performed according to the following decision rule: min ED(x^nj) where ED is the Euclidean distance value x the sample point to be classified, mi the nearest sample point of the ith class. After determinating the nearest neighbours K= 5, 7,... etc., the reliability of the classification M [%] i can be calculated using the equation: 2 i/EDßmax M[%] = 100 K 2 l/EDo e e=l where 2 l/ED£ max is the sum of the reciprocal values of Euclidean distances in that class to which the sample point mostly belongs, 2 l/EDp_ e is the sum of the reciprocal values of the Euclidean distances in all neighbours taken into account e = 1, K. A number of techniques exist which may be used effectively to estimate the required conclusions. It should be noted, however, that only the joint use of various pattern recognition methods provides reliable results whereas a single method can lead to erroleous conclusions.
