Abstarct: Quantitative structure–property relationship (QSPR) modeling is an efficient and cost-effective strategy for predicting chromatographic properties of chemical compounds baxsed on molecular structure. In this study, QSPR models were developed to predict the retention indices of selected polycyclic aromatic hydrocarbon (PAH) compounds using molecular descxriptors. A comprehensive set of descxriptors was calculated and subsequently reduced using stepwise regression (SR) and the firefly algorithm (FF) as variable selection techniques. Artificial neural networks (ANNs) were employed to model the potentially nonlinear relationships between molecular descxriptors and retention index. Model performance was assessed using statistical criteria, including the coefficient of determination and error-baxsed metrics for both training and external test sets. The results show that SR–ANN and FF–ANN models exhibit comparable coefficients of determination, indicating similar predictive accuracy. Notably, the SR–ANN model achieved this level of performance with a smaller number of descxriptors, resulting in a more parsimonious model with reduced complexity. These findings highlight the effectiveness of ANN-baxsed QSPR models and demonstrate that stepwise regression can provide an efficient descxriptor selection strategy without sacrificing predictive capability. The proposed approach offers a reliable tool for predicting the retention indices of PAH compounds and may reduce the need for extensive experimental measurements.