A feedforward neural network (NN) for archaeometric studies has been created to facilitate provenance attribution of archaeological ceramics. A multilayer perceptron model (MLP) has been applied to construct the network, including only one hidden layer. Moreover, correction parameters based on historical information has been applied to Bayesian probability factor. The NN has been trained by using clays mixings mathematically constructed using a reference chemical database of Sicilian sediments. The clay mixing take in consideration compositional variability within the same geological site and ceramic production processes. Test has been performed by querying the NN with compositional data of pottery assemblages found in the archaeological sites of Agrigento, Gela, Siracusa, Lentini, Catania and Milazzo coherently with clays sampling areas. Up to 88% correct attribution has been verified, with flawless correspondence between geological and archaeological contexts. Merits of NN have been finally highlighted, comparing the extent of successfully provisional attribution with results achievable by using classical multivariate statistical method (PCA and LDA).
Artificial Neural Network for the provenance study of archaeological potteries using clay sediment database
Raneri S.;
2018-01-01
Abstract
A feedforward neural network (NN) for archaeometric studies has been created to facilitate provenance attribution of archaeological ceramics. A multilayer perceptron model (MLP) has been applied to construct the network, including only one hidden layer. Moreover, correction parameters based on historical information has been applied to Bayesian probability factor. The NN has been trained by using clays mixings mathematically constructed using a reference chemical database of Sicilian sediments. The clay mixing take in consideration compositional variability within the same geological site and ceramic production processes. Test has been performed by querying the NN with compositional data of pottery assemblages found in the archaeological sites of Agrigento, Gela, Siracusa, Lentini, Catania and Milazzo coherently with clays sampling areas. Up to 88% correct attribution has been verified, with flawless correspondence between geological and archaeological contexts. Merits of NN have been finally highlighted, comparing the extent of successfully provisional attribution with results achievable by using classical multivariate statistical method (PCA and LDA).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
