Application of Decision Tree Algorithm for classification and identification of natural minerals using SEM-EDS

作者：Akkas Efe^*; Akin Lutfiye; Cubukcu H Evren; Artuner Harun

来源：Computers & Geosciences, 2015, 80: 38-48.

DOI：10.1016/j.cageo.2015.03.015

• Mineral classification
• Energy dispersive spectrometry
• Decision Tree Algorithm
• Scanning Electron Microscope
• C5.0 Decision Tree

A mineral is a natural, homogeneous solid with a definite chemical composition and a highly ordered atomic arrangement. Recently, fast and accurate mineral identification/classification became a necessity. Energy Dispersive X-ray Spectrometers integrated with Scanning Electron Microscopes (SEM) are used to obtain rapid and reliable elemental analysis or chemical characterization of a solid. However, mineral identification is challenging since there is wide range of spectral dataset for natural minerals. The more mineralogical data acquired, time required for classification procedures increases. Moreover, applied instrumental conditions on a SEM-EDS differ for various applications, affecting the produced X-ray patterns even for the same mineral. This study aims to test whether C5.0 Decision Tree is a rapid and reliable method algorithm for classification and identification of various natural magmatic minerals.
Ten distinct mineral groups (olivine, Orthopyroxene, clinopyroxene, apatite, amphibole, plagioclase, K-feldspar, zircon, magnetite, biotite) from different igneous rocks have been analyzed on SEM-EDS. 4601 elemental X-ray intensity data have been collected under various instrumental conditions. 2400 elemental data have been used to train and the remaining 2201 data have been tested to identify the minerals. The vast majority of the test data have been classified accurately. Additionally, high accuracy has been reached on the minerals with similar chemical composition, such as olivine ((Mg,Fe)(2)[SiO4]) and orthopyroxene ((Mg,Fe)(2)[SiO6]). Furthermore, two members from amphibole group (magnesiohastingsite, tschermakite) and two from clinopyroxene group (diopside, hedenbergite) have been accurately identified by the Decision Tree Algorithm. These results demonstrate that C5.0 Decision Tree Algorithm is an efficient method for mineral group classification and the identification of mineral members.

• 出版日期2015-7