Sensitivity of 2D Carbide Gas Sensor Materials: A Machine Learning (ML) Approach to Determine Most Influential Properties and the Best ML Model

Abstract The present study explores a series of machine learning models to assess the most influential properties for sensitivity of six different 2D carbide compounds of three classes, viz., X2C (Be2C), XC3 (PC3, AlC3, GaC3), and X2C3 (Sb2C3, Bi2C3), towards various environmentally toxic (NH3, NO2, NO, N2, CO, and CO2) and non-toxic (O2 and H2O) gases. Five popular machine learning (ML) methods, viz., logistic regression, random forest, naive bayes, support vector machine, and decision tree, are explored in the present study. The ML models are exercised to assess sensitivity in respect to eight different materials−gas interaction parameters predicted through density functional theory (DFT) investigation, including energy band gap (Eg), adsorption energy (Eads), charge density difference (Δρ), recovery time (τ), and work function (Φ). The ML studies reveal that the accuracy, F-measure (F1) score, and the area under the curve (AUC) values under random forest analysis are found to be 0.75, 0.70, and 0.85, respectively, which confirms it as the best among all the considered models. Therefore, the random forest model is proposed to be the most appropriate method in assessing the sensitivity of the considered 2D carbide series of materials.