Abstarct: Quantitative Structure-Property Relationship (QSPR) is one of the efficient and powerful methods for predicting the physicochemical properties of compounds baxsed on their molecular structure. In this study, QSPR models were developed to predict the retention time (RT) of phytochemical compounds in Bingöl honey bee pollen. The molecular structures of the compounds were drawn with ChemDraw and Avogadro software and optimized with the semi-empirical AM1 method. Then, using Dragon software, thousands of molecular descxriptors (zero-dimensional, one-dimensional, two-dimensional and three-dimensional) were calculated and after preprocessing (removal of constants and high collinear variables), they were selected with three advanced methods: Stepwise Regression (SW), Firefly Algorithm (FF) and Random Forest (RF). The selected descxriptors were mainly from the 3D-MoRSE, GETAWAY, Constitutional and Topological groups, covering electronic, geometric, topological and molecular charge distribution features. The performance of the models was evaluated with statistical parameters, external test set, scatter plot, residual plot, applicability domain (Williams Plot with AD > 96%) and Y-random test (average R² random ≈ 0.001). The RF-ANN and FF-ANN models showed the best performance in the external test set and LOO and AD evaluations; high R² and wide applicability domain confirmed the prediction accuracy. The SW-ANN model also had acceptable performance but provided less accuracy than the group and nonlinear models in complex nonlinear relationships.