Neural network analysis for predicting metrics of fragmented laminar artifacts: a case study from MPPNB sites in the Southern Levant

Abstract

Abstract This study was aimed at introducing a new method for predicting the original metrics of fragmented standardized artifacts, specifically of flint blades from the Middle Pre-Pottery Neolithic B (10,200/100–9,500/400 cal B.P.) in the Southern Levant. The excessive re-use of these artifacts or poor preservation conditions often prevent a complete set of metric data from being obtained. Our suggested approach is based on readily accessible machine learning (artificial intelligence) and neural network analysis. These are performed in a multi-paradigm programming language and numeric computing environment, with parameters represented by a rapid measurement system based on the technological features shared by all lithic artifacts of the studied assemblages. This method can be applied to various chronologies and/or contexts. A full set of metric data, including potential typological and functional elements of the assemblages studied, may provide a better understanding of the lithic technology involved; highlight cultural aspects related to the chaîne opératoire of the studied lithic production; and address issues related to cultural sub-divisions in larger-scale applications. Herein, neural network analysis was performed on blade samples from Middle Pre-Pottery Neolithic B sites from the Southern Levant specifically Nahal Yarmuth 38, Motza, Yiftahel, and Nahal Reuel.

Keywords

• Neural network analysis
• Machine learning
• Metric prediction
• Lithic industry
• Pre-pottery neolithic B
• Southern Levant