Accurate prediction of gas hydrate formation temperature is essential for ensuring flow assurance during natural gas production and subsea transportation. In this study, 459 experimental data points collected from the literature, covering a wide pressure range of 330-68,600 kPa and gas specific gravity range of 0.552-1.03, were utilized to develop reliable predictive tools using only two practical input variables: pressure and gas specific gravity. Two complementary modeling approaches were implemented. First, a systematic comparative analysis of several advanced machine learning technique, including Multilayer Perceptron (MLP) trained with Levenberg – Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) algorithms; Radial Basis Function (RBF); CatBoost; XGBoost; and Least-Squares Support Vector Machine (LSSVM), was conducted. Second, an optimized empirical correlation was formulated, with its coefficients determined using the Generalized Reduced Gradient (GRG) nonlinear optimization algorithm to minimize overall prediction error across the entire experimental dataset. Among the machine learning models, CatBoost demonstrated superior predictive performance, achieving an AAPRE of 0.094 % and an R2 value of 0.996 on the testing dataset. The developed GRG-based correlation also exhibited strong agreement with experimental data while providing a transparent and computationally efficient alternative suitable for rapid engineering calculations. Sensitivity and leverage analyses confirmed the robustness, statistical reliability, and broad applicability of the proposed models within the investigate domain.
