BPSO-MLR and ANFIS based modeling of lower flammability limit

Computer-aided molecular design (CAMD) is a methodology that has been researched recently for designing materials with required target properties and can be well applied for the selection of safer chemicals in chemical process design. An important aspect of this methodology concerns quantitative structure property relation (QSPR) based studies. The main objective of this work is to provide simple straightforward techniques for the accurate prediction of the lower flammability limits through a robust QSPR approach. The obtained three-parameter multivariate regression (MLR) and adaptive neuro-fuzzy inference system (ANFIS) modeling strategies resulted in encouraging statistics of R2 = 0.906, RMSE = 0.335 and R2 = 0.930, RMSE = 0.287 vol.%, respectively, using the entire DIPPR dataset. Using such a large dataset comprising of 1615 compounds from 82 diverse chemical material classes can highly benefit the study accuracy and comprehensiveness. Notably, the model input variables can be easily calculated for non-specialist user.

Simple yet accurate estimation of lower flammability limit for 1615 organic and simple inorganics (RMSEall = 0.335).Figure optionsDownload as PowerPoint slideHighlights
► LFL of 1615 compounds comprising 82 diverse chemical material classes were studied.
► The MLR resulted to simple yet accurate model with practical application in ISD.
► Nonlinear behavior of LFL was considered using an optimized ANFIS modeling tool.
► The models showed higher generalizability and accuracy with simple input variables.