Recently, air traffic demand has been increasing rapidly which creates saturation, congestion and queues issues. Flight delays put an immense load on transportation networks, resulting in operational shortcomings, higher economic losses for airlines and complaints and non-satisfaction for their customers. For this reason, aviation authorities, academics and researches become more interested in taking preventive actions in order to minimize the effect of flight delays. In this paper and based on a questionnaire distributed among pilots, air traffic controllers, airport administrators, airport staff, and travelers, we proposed a predictive model in order to forecast traffic delays for United States domestic flights. Four effective Machine Learning algorithms such as feed-forward Neural Network, Gradient Boosting, Decision Trees and Random Forest were employed on a data from 1st January to 31st December, 2018. We extracted from the dataset the traditional features that are contributing to flight delays. To enhance the precision of the suggested model, we created and proposed new features that, the best of our knowledge, were never treated in existing researches. Hence, the model was capable to predict flight delays with a higher coefficient of determination of 98.70% in the case of feed-forward neural network with a lower error value when both traditional and suggested features were utilized.