MODELLING OF PICRALIMA NITIDA EXTRACTION AND OPTIMIZATION OF THE EFFECT OF PROCESS PARAMETERS USING ARTIFICIAL NEURAL NETWORK AND RESPONSE SURFACE METHOLOGY.

ABSTRACT

Picralima nitida is a deciduous tree that finds its utilization in pharmacology and tradomedicine. It has been investigated to contain certain bioactive components such as alkaloids,flavonoids, tanins and saponins. In this current study, the bioactive components were extracted using ethanol as solvent under various process parameters; particle sizes, temperatures and extraction times. The extraction process was optimized in order to get the changes in the process parameters that will give us the optimum yield. It was found that there is a direct relationship between extraction yield and temperature, a direct relationship between extraction yield and time, and also found that the relationship between extraction yield and particle size was an inverse on. Optimizing and modelling the extraction process using two optimization tools; artificial neural network and response surface methodology. RSM was used to obtain a quadratic model for predicting the extraction yield keeping particle size, temperature and extraction time as the varying parameters. Also ANN was used to optimize and model, Bayesian regularization learning algorithm with the hyperbolic (Tanh) for the hidden and the output layers was the best model for the predictive modelling of the extraction yield. The performance of both models was established based on R2 and RMSE values. While ANN gave R2 value of 0.97708 and RMSE value of 0.063578, RSM gave an R2 value of 0.9296 and RMSE value of 2.07.Also, while RSM gave an optimum yield of 42.87 at 90o c, 50mins and 1mm temperature, extraction time and particle size respectively, ANN gave a better optimum yield of 40.5762 at the same conditions. From the predicted yields given by both models, it is seen that RSM gave a higher optimum predicted yield, hence RSM should be selected as a better modelling and optimization tool.