Analysis and prediction of performance indicators of Brazilian air traffic by machine learning

The Brazilian air transport sector plays a significant role in the country's economy. For example, in 2018, this sector contributed 1.1% to the Gross Domestic Product (GDP) and generated 839,000 direct jobs. Although the industry was severely affected by the pandemic, growth rates are gradually returning, making it essential to analyze the performance of airports and airlines. One way to perform this analysis is through performance indicators such as the Key Performance Indicators (KPIs) of the International Civil Aviation Organization's (ICAO) Global Air Navigation Plan (GANP). This article analyzes departure punctuality (KPI 01) and identifies the key factors affecting punctuality to improve airport and airline management. To conduct this analysis, the Extreme Gradient Boosting (XGBOOST) computational algorithm was used to predict whether a flight would be classified as punctual or not. Additionally, the XGBClassifier algorithm was used to assess the input variables' importance in the supervised machine learning model. The variables employed in the machine learning analysis include local meteorological variables at the time of departure, operational data on aircraft demand and capacity, specific geolocation information, and temporal data collected from public databases. The XGBoost model used to evaluate the punctuality of Brazilian airports achieved an accuracy of 92.96\%. For the analyzed airlines, the departure accuracy rates were 91.6% for GOL, 93.9% for TAM, and 94.0% for AZUL. Furthermore, the results indicated that atmospheric pressure, the number of flights departing simultaneously, and the weight of checked baggage were the main factors affecting airport departure operations. For airlines, the variables of atmospheric pressure, free baggage, and paid baggage had the most significant impact on punctuality.