Determination of equilibrium composition for various multi-phase systems is important in the context of thermodynamics. Three methods are generally employed to calculate the gas/liquid equilibrium compositions; namely, empirical graphs, correlations, and equations of state (EOSs). Empirical graphs and correlations are simple and fast in terms of calculation procedure. Furthermore, using an EOS requires an initial guess, which is usually obtained via empirical correlations. In this study, the gas-oil composition of 10 different crude oils (20-40 degrees API) are experimentally determined by a gas chromatography (GC) apparatus within a temperature range of 600-1212 degrees R and a pressure range of 14.7-7000 psi. A robust predictive model is then proposed to estimate the equilibrium ratios (K-i) of hydrocarbons and non-hydrocarbons. This model is generated by utilizing the least squares support vector machine (LSSVM), while genetic algorithm (GA) is used for selection and optimization of hyper parameters (gamma and sigma(2)) that are embedded in the LSSVM model. The coefficient of determination (R-2) for the introduced model is 0.9991 and 0.9979 and the mean squared error (MSE) is 0.00074 and 0.044 for the hydrocarbons and non-hydrocarbons, respectively. The proposed model is simple to use and exhibits high accuracy and reliability, which can have various applications in chemical and petroleum industries where the thermodynamic equilibrium is maintained.