Abstarct: Quantitative Structure-Property Relationship (QSPR) is one of the efficient methods for predicting chromatographic parameters of organic compounds baxsed on their molecular structures. In this study, QSPR models were developed to predict the Retention Index (RI) of a set of non-polar and low-polarity organic compounds on standard non-polar stationary phases (polydimethylsiloxane). To extract molecular descxriptors, the three-dimensional structures of the compounds were first optimized using molecular mechanics, semi-empirical (AM1), and density functional theory methods in HyperChem software. Subsequently, the descxriptors were calculated using Dragon software. Four advanced methods were employed for optimal descxriptor selection: Stepwise Multiple Linear Regression (Stepwise MLR), Ant Colony Optimization (ACO), Random Forest (RF), and Genetic Algorithm (GA). Nonlinear modeling of the relationships between the descxriptors and retention indices was performed using Artificial Neural Networks (ANN). The models were evaluated using multiple statistical criteria (R², R²adj, MSE, Q²_LOO), Y-randomization test (1000 iterations), Williams plot, and external validation parameters (R²₀, R²m, K, and K'). The results demonstrated that the GA–ANN and RF–ANN models provided high predictive accuracy on the external test set; however, the ACO–ANN model, with a smaller number of descxriptors (5 variables), achieved an excellent balance between accuracy, simplicity, and interpretability. The Y-randomization test yielded mean R² values below 0.002 for all methods, confirming the non-random nature of the models. Furthermore, the Williams plots indicated an applicability domain coverage exceeding 93.8% without significant outliers. These findings confirm the high efficiency of the combined variable selection and neural network approach for predicting retention indices. The proposed models offer a fast, cost-effective, and reliable tool for estimating the RI of new organic compounds, significantly reducing the need for extensive experimental measurements in GC-MS analyses.