Abstarct: Chocolate is one of the most widely consumed food products in the world, which has a special place in the diets of people all over the world. Volatile and semi-volatile organic compounds (VOCs) present in dark chocolate are the main factor in creating its special and distinctive aroma. The aroma consists of a complex set of volatile compounds. In the present study, QSPR models were built with the aim of predicting the inhibition index of some volatile and semi-volatile organic compounds (VOCs) present in dark chocolate, using molecular descxriptors. For this purpose, first, the number of calculated descxriptors was reduced by using stepwise regression (SW), genetic algorithm (GA) and ant colony algorithm (ACO) as variable selection techniques. Artificial neural networks (ANN) were used to establish a relationship between molecular descxriptors and the inhibition index - which may have nonlinear relationships. The performance of the models was evaluated using various statistical measures, including coefficient of determination and error-baxsed parameters, for the external training and test sets. The results showed that the SR-ANN, ACO-ANN, and GA-ANN models had close coefficients of determination, indicating almost identical predictive power. However, the ACO-ANN model was able to achieve this level of accuracy using a smaller number of descxriptors; as a result, a simpler, less complex, and more computationally efficient model was presented. The proposed method provides a reliable tool for estimating the inhibition indices of the studied compounds and can significantly reduce the need for extensive and costly experimental tests.