Comparison of ANN & RSM approaches for optimum production of Y-Decalactone

Y-Decalactone is an important flavor compound widely used in food, diary and fragrances. Previously it was produced from fruits through chemical synthesis. Due to increase in demand for natural products by consumers, it has gained interest for its production through biotechnological way. The present study focusses on fermentation variables optimization for the production of Y-decalactone using Sporidiobolus salmonicolor via Response surface methodologies (RSM) and Artificial neural networks (ANN) using castor oil as substrate. The prediction abilities of RSM and ANN were compared based on error parameters namely Mean absolute error (MAE), Root mean square error (RMSE), chi-square (x²) and correlation coefficient (R²) to suggest the best approach for modeling. The response variable (Y-decalactone production) was modelled and optimized as a function of four input variables (castor oil percentage, pH, incubation time and temperature). Training of ANN network was performed using a multilayer feed forward architecture with same experimental datasets used in RSM. Model predictions of both approaches were compared with the experimental values and reported that these are in good close agreement. The highest production of Y-decalactone 72.73 mg/l obtained at an optimum conditions of castor oil −29.68%, pH −5.32, incubation time −99.89 h and temperature −23.22⁰C. Hence, results were beneficial in using appropriately trained ANN over RSM for nonlinear fermentation systems.