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

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

reaching the surface of glass as a result of electrochemical processes, and then diffusing into the bulk of glass or vice versa. The results of these investigation allow archaeologists to answer important questions, e. g. the manufacturing technology of glass objects, or to determine whether the object was manufactured at the given site or whether it was imported. 1 3 5 7 1 3 5 7 —0,2mm measuring positions Ffc.2 Concentration distribution on the surface and in the bulk of glass samples Classification of glasses by M VDA For the determination of connections between the quality of artificial products on the one hand and analytical data on the other hand multivariate statistical methods become increasingly important. Analytical investigations frequently serve to clarify nonanalytical questions. This means that qualitative decisions have then to be made from quantitative analytical data. The interpretation of the data required for this has become more and more difficult over the past few decades. An interpretation of multi-dimensional analytical data analogous to the classical way by means of repeating the univariate operations many times is not recommended [29]. A solution for multivariate problems can be achieved mathematically, Pattern Recognition Methods having been developed specifically for this aim. For classifying data and plotting results so-called non-elementary discriminant functions representing linear combinations of the original data are used. For the actual data classification, supervised learning and unsupervised learning methods are principally distinguished. In Supervised Learning, the number of classes is given (normally for self-evident reasons) for which one determines discriminant functions in so-called learning and training phase programmes. With the help of this unknown objects can be classified into the given classes during the actual working phase. The separation of the classes from one another is achieved by means of 95% confidence level.