A study of ancient glass type prediction based on logistic regression models

Abstract

Ancient glasses identification and classification has been widely developed to investigate the ancient culture and historical issues. However, existing researches are concentrated on the classifications for ancient glass with unacceptable identification accuracy, which ignores the prediction method can also successful identify the types of ancient glass. Therefore, there is a demand for correctly predicting types of ancient glass with acceptable identification accuracy and reasonable system costs. Logistic regression models are statistical methods used to predict a binary (yes/no, 0/1) outcome based on one or more predictor variables. The aim of this paper is to make species predictions for ancient glass based on chemical composition distributions, and to determine whether an unknown glass sample is a high potassium or leadbarium glass. We collected data from 61 sets of samples, each corresponding to fourteen chemical compositions, and used a logistic regression algorithm to make species predictions for these sample glasses, solving for the parameters using Newton's method and gradient descent, and comparing them.