Oxide’s Acidity Prediction in Smith Scale Using a Simple Linear Model

A numeric scale of acidity and basicity, developed by D.W. Smith, allows for quantitative comparison between different oxides and has been useful in explaining various oxide behaviors. In this study, we aim to predict oxide acidity on the Smith scale using a machine learning approach. Previous attempts using a linear fit based on electronegativity showed a clear trend but lacked precision due to the simplicity of the model and the multi-valent nature of metal oxides. We propose a multi-parameter model incorporating four features: electronegativity, metal valence (raised to the power of 1/3), ionic radius, and dipole polarizability. A simple linear neural network outperformed a more complex 1D convolutional neural network, demonstrating superior accuracy and interpretability. Our model significantly improved predictive performance as compared to the single-parameter model. These results were validated through predictions of testing data and unknown oxides acidity were predicted in this work.