Resumen
A multilayer feedforward neural network (MFNN) model is developed from the best performing backpropagation training algorithm among all eleven training algorithms. From experimental setup of microwave pyrolysis of oil palm fibre, the model utilized the input data which are microwave power, temperature and Nitrogen flow rate whereas the output data are weight of hydrogen and biochar. Eleven training algorithms employed belong to six classes which are (1) additive momentum, (2) self-adaptive learning rates, (3) resilient backpropagation, (4) conjugate gradients, (5) quasi-Newton and (6) Bayesian Regulation. To compare the performance of different training algorithms, error values are estimated by performance function root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). The objective of this research is to identify the most suitable training algorithm for this process and it is found that Bayesian Regulation has the best performance with MAPE values of 0.0119 and 0.0131 for weight of hydrogen and weight of biochar respectively.