Classifying Crystal Structures of Binary Compounds AB through Cluster Resolution Feature Selection and Support Vector Machine Analysis

• DOI: 10.1021/acs.chemmater.6b02905
• Publication Date: 2016-08-22T16:42:37Z
• Authors: Anton O. Oliynyk; Lawrence A. Adutwum; James J. Harynuk; Arthur Mar
• Keywords: Optimization; Informatics; Automotive Gasoline Samples; Scales; Principal components analysis; 02 engineering and technology; Ionization-Potentials; Classification; Electronegativity; Predictions; Neural-Networks; 0210 nano-technology

Partial least-squares discriminant analysis (PLS-DA) and support vector machine (SVM) techniques were applied to develop a crystal structure predictor for binary AB compounds. Models trained validated on the basis of classification 706 compounds adopting seven most common types (CsCl, NaCl, ZnS, CuAu, TlI, β-FeB, NiAs), through data extracted from Pearson’s Crystal Data ASM Alloy Phase Diagram Database. Out 56 initial variables (descriptors based elemental properties only), 31 selected in as unbiased manner possible procedure forward selection backward elimination, with quality model evaluated by measuring cluster resolution at each step. PLS-DA gave sensitivity 96.5%, specificity 66.0%, accuracy 77.1% validation set data, whereas SVM 94.2%, 92.7%, 93.2%, significant improvement. Radii, electronegativity, valence electrons, previously chosen intuitively maps, confirmed important variables. could also make quantitative predictions hypothetical compounds, unlike semiclassical approaches. The new compound RhCd was predicted have CsCl-type (0.669 probability) and, an even stronger confidence level, (0.918 probability). synthesized reaction elements 800 °C X-ray diffraction adopt structure. is thus superior method crystallography that fast makes correct, predictions; it may be more broadly applicable help identify unknown any arbitrary composition.