Process modeling of solvent extraction of oil from Hura crepitans seeds: adaptive neuro-fuzzy inference system versus response surface methodology

Abstract

Vegetable oils are a very important feedstock for many industries such as biofuels. There is the need to source for novel and underexploited plant oilseeds to meet the world demand for oils. Thus, the extraction of oil from Hura crepitans (sandbox) seeds was conducted using the solvent extraction method. Modeling of the extraction process was carried out using response surface methodology (RSM) and adaptive neuro-fuzzy inference system (ANFIS). The effects of the nature of the solvent (non-polar (n-hexane) and polar (acetone and ethyl acetate)), solid-solvent ratio (0.1–0.3 g/mL), extraction time (2–6 h), and their interactions on the oil yield were investigated using the D-optimal design technique. Performance assessment of the developed models was carried out to check their effectiveness in predicting the H. crepitans seed oil (HCSO) yield using various fit statistics. The coefficient of determination (R2) observed for the RSM and ANFIS models was 0.9720 and 0.9988, respectively, with corresponding mean relative percent deviation (MRPD) of 2.50 and 0.37%. Maximum HCSO yield of 62.95 wt% was achieved by ANFIS coupled with genetic algorithm (GA) using 0.1 g/mL solid-solvent ratio, extraction time of 4.19 h, and acetone, while maximum HCSO yield of 62.50 wt% was observed by RSM with a solid-solvent ratio of 0.1 g/mL, extraction time of 4.04 h, and acetone. Characteristics of the HCSO indicated that it could serve as a good feedstock for the production of oleochemicals such as biodiesel. The results obtained in this study demonstrated that ANFIS is marginally superior to RSM in the modeling of the HCSO extraction process, while GA was slightly better than the numerical tool of RSM in the optimization of the process.