Abstarct: The purpose of this project is to use the artificial neural network and random forest modeling method, to select the best descxriptors using the fireFFy algorithm and the particle swarm optimization algorithm. FF and PSO variable selection method with the most selected descxriptors, the number of 10 and 6 descxriptors (for both selection methods) that are most related to the pharmaceutical activities of the selected compounds. Then the selected descxriptors were used as input of artificial neural network and forest to create QSAR models. For the neural network model, the data series were divided into training, evaluation and test series. For the random forest model, the data series were divided into two parts, training and test series. In order to build the optimal neural network model with proper efficiency, all counseling centers in the network were trained optimally. After creating the optimal model, the pharmaceutical activity compounds of the test set were predicted. The coefficient of determination for test data in FF-RF, PSO-RF, FF-ANN and PSO-ANN was equal to 0.78, 0.67, 0.81 and 0.8. The developed models (FF-RF, PSO-RF, FF-ANN and PSO-ANN) are evaluated using different statistical methods such as single-stage rejection tests, domain of application, dispersion, Y-random and calculation of a series of statistical measures. The evaluation results prove the generalizability and predictive power of the developed PSO-ANN and FF-ANN models.