Prediction and interpretability analysis of the properties of energetic ionic salts based on machine learning

As a significant subset of energetic materials, energetic ionic salts (EISs) have garnered considerable attention due to their potential application in the fields of explosives and propellants. To accelerate their development, this study establishes machine learning models for predicting key properties of EISs including density (<i>ρ</i>), enthalpy of formation (Δ<i>H</i>_f), detonation velocity (<i>D</i>_V), detonation pressure (<i>P</i>), and thermal decomposition temperature (<i>T</i>_d). Using a customized descriptor set (CDS) and a dataset of 1202 EISs, five ML algorithms were evaluated. The results show that the optimal prediction model varies depending on target properties: MLP performs best for <i>ρ</i> (<i>R</i>² = 0.91 and MAE = 0.036 g cm^-3), SVR for <i>P</i> (<i>R</i>² = 0.73 and MAE = 1.86 GPa), KRR for Δ<i>H</i>_f (<i>R</i>² = 0.85 and MAE = 103.24 kJ mol^-1) and <i>D</i>_V (<i>R</i>² = 0.69 and MAE = 230.97 m s^-1), and GBR for <i>T</i>_d. Prediction accuracy for <i>ρ</i> and Δ<i>H</i>_f exceeded that for detonation properties and <i>T</i>_d. Submodels based on nitro group attachment (C-NO₂<i>vs.</i> N-NO₂) improved <i>T</i>_d prediction. SHAP analysis revealed that oxygen balance, and hydrogen and nitrogen counts are universally important, alongside property-specific descriptors. This work demonstrates the potential of ML in predicting the key properties of EISs, thereby accelerating their structural design.