AdaBoost Metalearning Methodology for Modeling the Incipient Dissociation Conditions of Clathrate Hydrates.

This paper proposes the AdaBoost metalearning methodology to combine the outcomes of tree-based models of classification and the regression tree (CART) algorithm for estimating the equilibrium dissociation temperature of clathrate hydrates. In addition to the AdaBoost-CART models, models based on the adaptive neuro-fuzzy inference system (ANFIS) and artificial neural network (ANN) approaches were also developed. Training and testing of the models were done utilizing a gathered database of more than 3500 experimental data on incipient dissociation conditions of CO and other hydrate systems. With the average absolute relative deviation percent (AARD%) between 0.03 and 0.07, 0.04 and 1.09, and 0.09 and 1.01, which were obtained by the presented AdaBoost-CART, ANFIS, and ANN models, respectively, the targets were reproduced with satisfactory accuracy. However, for all of the studied clathrate hydrate systems, the proposed AdaBoost-CART models provide more reliable results. Indeed, the obtained AARD% values for tree-based models are lower than those of other models.