A datamining approach to classify, select and predict the formation enthalpy for intermetallic compound hydrides

Abstract In this paper, two techniques of datamining tools were adopted, a principal component analysis (PCA) and artificial neural network (ANN). A PCA to classify, select and identify several combinations between transition element A and B (B = Ti, Zr, Hf, Sc, Y, La and Th) and ANN to predict Δ H for ternary hydrides. Based on the datasets selected from different works, a principal component analysis (PCA) has been applied to select, classify and identify around 76 possible combinations between transition metal elements A and B. The results showed that the clustering of combinations A-B are significantly influenced by the atomic parameters of element A, such atomic radius ( R A ), Pauling's electronegativity (χ A ) and atomic electron density ( Z A / R A 3 ). From 76 combinations, 55 systems which have χ A  ≥ 1.5, Z A / R A 3 >1.28 and R A A Z A / R A 3 R A  > 1.46 A are categorized as group 2. From the first group, 46 different combinations are identified and have a negative Δ H , within 18 well-known promising binary alloys of hydrogen storage. An (6-15-1) architecture of artificial neural network (ANN) has been developed to estimate the Δ H for the other ternary hydrides selected from different published works. The performance indices such as relative error, coefficient of determination ( R 2 ) and mean square error (MSE) were used to control the performance of obtained results. In addition to this, the Δ H obtained from ANN model were compared with those experimental data and theoretical results available in the literature.