Using machine learning algorithms to predict the risk of small Unmanned Aircraft System violations in the National Airspace System

The increasing number of small Unmanned Aircraft System (sUAS) encounters with manned aircraft or airports increases the risk of collision in the National Airspace System. The purpose of this research is to develop and test predictive models for sUAS violation incidents in NAS using machine learning. This research uses machine learning algorithms to predict the risk of sUAS violation incidents using the FAA's UAS sighting data with a sample size of 2088. Three sUAS violation types are identified: flying above 400 feet, flying with 5 miles from an airport, and flying in restricted airspace. Seven machine learning algorithms were used, including classification regression, decision tree, neural network, gradient boosting, random forest, Bayesian networks, and Memory-Based Reasoning. The results show that Gradient boosting produces the best predictive model. This model can predict the sUAS violation incidents with an accuracy of 95.7 percent. Location, distance to the airport, state, sUAs altitude, airport type, and aircraft type are the most influential predictors to the sUAS violation incidents.