M. Járó - L. Költő szerk.: Archaeometrical research in Hungary (Budapest, 1988)

Analysis - ZIMMER Károly: Spectrochemical investigation and classification of Hungarian glass finds

With Unsupervised Learning Methods the number of classes is not given, but an attempt is made to find and to form classes on the basis of the relations of multidimensional similarities between the objects. Applications. Historical glass bead findings, by virtue of their spectroscopically determined trace contents of colouring elements, can be classified by supervised learning method. The results of discriminant analysis are often not unambiguous owing to the frequent overlap of variance regions if all samples representing the eight colour classes are taken into account. The outliers and the inclusion of "mixed" class give most probably false class means and increase intraclass variances Therefore, the same process was repeated with the omission of some objects. Thus, 6 classes were determined for 33 objects. The necessary parameters were determined again (class mean, intraclass variance, means and variances pertaining to the 13 elements). Owing to the little variation of nickel content, this element was omitted. All objects were classified correctly into the a priori class, and only one overlap of variance regions occurred — between the violet and blue classes as can also be seen in Fig. 3. The objects omitted according to the above were included now in the resulting classes. Some of the objects fall outside the variance limits. Multivariate variance and discriminant analysis proved to be excellent methods in the cases investigated for the classification of glass fragments on the basis of their colourants if the outliers recognizable by univariate methods are omitted before their application [28]. This, on the basis of our knowledge on glass colourants, according to ligand field theory, is valid only if glasses prepared from the same basic materials by the same process are compared. Therefore, it is plausible to assume that the glasses come from the same age, in which the method of manufacture had not changed significantly and the place of production was also the same. This statement is questionable for some of the objects which fell outside the given variance ranges. The classification of archaeological glass finds to their origin (Byzantine, Venetian, German, Hungarian, and made in Hungary according to Venetian models) is a more difficult problem [30]. Although the classes widely overlap (because the trace element pattern alone is not significant for the origin), it was possible to classify 66 out of 69 samples as being correct or plausible from the archaeological point of view. The classification results by MVDA are represented in Fig. 4 and Table 2. Table 2 Qualification of glass samples by origin —Qualification Origin Reliable origin Correct Corrected Not accepted New Reliable origin qualification Byzance Germany Venice Hungary Hungary (Venetian style) 5 1 8 1 2 3 2 3 7 2 1* 2 6 7 6 13 Number of samples 15 7 12 32 * The sample was probably made in another country