Optimisation of neural network topology for predicting moisture content of spray dried coconut milk powder / Nadiah Syafiqah Shaharuddin, Zalizawati Abdullah and Farah Saleena Taip

Moisture content is an important parameter to be controlled in an agricultural product to prevent rapid degradation and promote stabilisation of product and longer shelf life. There are various techniques on predicting the moisture content of spray dried powder that had been used by past researchers. In this study artificial neural network (ANN) is proposed to be used for its well-known benefits of simplicity and accuracy. The aim of this research is to evaluate the effect of hidden layer and hidden neuron in ANN in predicting moisture content of spray dried coconut milk. The effect of training algorithm, e.g., Gradient Descent (GD) back propagation and Levenberg-Marquart (LM) back propagation, and activation functions, e.g., hyperbolic tangent sigmoid (tansig) and log sigmoid (logsig) functions are also studied. Based on the result of correlation coefficient of determination (R2) value of 0.9951 and root mean square error (RMSE) value of 0.0145 that was used to evaluate the ANN performance, it can be concluded that the best ANN topology is 2-10-1 with Levenberg-Marquart for learning algorithm, and tangent sigmoid as activation function.