Determination of pure alcohols surface tension using Artificial Intelligence methods

摘要

Reliable Artificial Intelligence methods are developed here for the determination of the surface tension of alcohols in a wide range of temperatures. A total amount of 3063 data was considered for 149 alcohols. Three methods were applied: multilayer perceptron neural network, radial basis neural network and, least-squares support-vector machine (LSSVM). Three different optimization algorithms were used for the multilayer perceptron and the support-vector machine, whereas only one was needed in the case of the radial basis neural network. So, seven different Artificial Intelligence methods were considered, trained and tested. Combination of 9 different input properties, their inverses, and their natural logarithms were studied, and the final input set chosen includes 12 of them. The obtained results were widely studied by using both graphical and common statistical tools. The applicability domain was obtained and different sensitivity analyses were performed.