Abstract: ‍ ‍Different from most of the current flight delay prediction methods that only tend to study a single classification or regression task， a flight delay prediction model based on multi-task NR-DenseNet was proposed in this paper， which aims to achieve both flight delay level classification prediction and delay time regression prediction. Firstly， the relevant data was pre-processed； Secondly， multi-task learning feature extraction sharing layer was established， NR-DenseNet was used to extract shared parameters between tasks to dig deeper into relevant features between tasks； Then， a multi-task learning task-specific layer was established to output the prediction results of specific tasks through a regressor and a classifier respectively； Finally， the loss-weighting method was used to optimize the loss functions of the two tasks， balance the convergence rate between tasks， and improve the generalization of the model. Compared with the single-task model， the average MSE of the regression task was reduced by 23.4%， the average MAE was reduced by 14.2%， and the average classification accuracy was increased by 2.7%. Experimental results show that the proposed method improves the accuracy of the classification task and reduces the error of the regression task， which can effectively improve the model performance.