Umudike Journal of Engineering and Technology

The study aimed at modelling of soft sensor models; SVM and GPR, for predicting the thermal property (specific heat) of dried turmeric rhizome for easy determination of the direct technical properties of the rhizome. Proximate composition analysis was conducted for each of the dried samples of turmeric to determine the nutrition composition. Dried turmeric thermal properties were derived from empirical equation for specific heat. Two hundred and ninety five (295) data set was used in developing, training and testing the models using fivefold cross validation method, five (5) of the remaining data was set aside used for independent validation of predictive model results. The models developed are Support Vector Machine (six variants) and Gaussian Process Regression models (four variants). The four input parameters time, temperature, air velocity and relative humidity were all simulated using both soft computing models. The result of the models indicated that Square exponential model of the GPR models have the best model convergence with the combination of all the input variables.