optimizing rubber seed oil extraction for biodiesel production using machine learning tools: a comparative study of response surface methodology and artificial neural networks

The efficiency of extracting oil from oil-bearing seeds is significantly affected by various process conditions, making optimization essential. this study utilizes the box-behnken design (bbd) to examine how solvent volume, sample weight, and particle size influence rubber seed oil yield during batch-mode solvent extraction using n-hexane. the optimization process was carried out using both response surface methodology (rsm) and an artificial neural network (ann). a quadratic model developed through rsm estimated the oil yield based on these key factors. for ann modeling, the optimal structure was identified as a multilayer full feed forward (mfff) network trained using the quick propagation (qp) learning algorithm. the hyperbolic tangent (tanh) function served as the best activation function for both hidden and output layers. the ann architecture included three input neurons, three hidden neurons, and one output neuron. according to the rsm model, the highest predicted oil yield was 56.57% under the conditions of 294.47 ml solvent volume, 10 g sample weight, and 1 mm particle size. meanwhile, the ann model estimated a maximum yield of 55.46% with a solvent volume of 300 ml under similar conditions. a comparative assessment revealed that ann performed better than rsm, achieving a higher coefficient of determination (r² = 0.9998) and a lower root mean square error (rmse = 0.3050), whereas rsm resulted in r² = 0.9789 and rmse = 0.7035. these findings indicate that ann provides superior accuracy and reliability in modeling and optimizing the impact of process parameters on rubber seed oil yield.