Airline rescheduling can be used to minimize the number of abnormal flightsduring the flight-plan execution process. This can improve the rate of normalflights and reduce the occurrence of flight delays and their subsequent adverseeffects. Airline rescheduling involves few samples and is also uncontrollable. Thesubjective manual reduction method has low efficiency and does not considerdaily major security and epidemic factors. Traditional machine learning methodsrequire massive data training and testing, and so are not applicable to the relativefew samples available for airline rescheduling. In this study we addressed theseproblems by establishing a set of airline rescheduling indicators by analyzing factors that influence airline rescheduling, and determined the inputs for an airlinerescheduling identification model based on a multilayer perceptron model. TheMAML (model-agnostic meta-learning)-SGD (stochastic gradient descent) airlinerescheduling algorithm was proposed to enable the identification model to learnand classify flights using few samples. By using historical data for GuangzhouBaiyun International Airport from May to September 2021 to train and testthe model, the airline rescheduling accuracy of the model reached 97% usingthis algorithm, which was markedly higher than that achievable using traditionalmachine-learning models such as SVM(support vector machines). Compared withthe identification model that uses an SGD algorithm to iteratively update parameters, this algorithm can maintain the stability of the gradient descent during thetraining and testing processes of the model.