Evaluation of Machine Learning Result for Metal Identification

Abstract

The purpose of this study is improvement of metal identification performance with step response. Feature values are maximum derivative current and its reaching time, these values depend on lift-off in the range of 0.5-1.5 mm. As a result of metal identifications, decision tree is the fastest and highest accuracy in 4 machine learning models. When increasing training samples, calculation time of all models are increasing, and accuracies are saturated 100 samples. When comparing between data whose lift-off is from 0.5 to 1.5 mm and data that fixed lift-off, classification accuracy in data fixed lift-off is improve than one in data not fixed lift-off.