The invention relates to the field of traffic data analysis, in particular to a subway passenger congestion degree prediction method adopting a resampling recurrent neural network. The method comprises the following steps: 1, carrying out pretreatment; training sample data are set according to the original data; setting crowding degree label, dividing the sample data into n sub-sample sets according to the congestion degree labels, resampling the sub-sample sets to obtain resampling sequences, inputting the resampling sequences into a recurrent neural network model to train the recurrent neural network model, evaluating the recurrent neural network model, and adjusting resampling weights according to an evaluation result until the evaluation result is good. In the prior art, random sampling is often carried out from training sample data, but different types of samples are distributed unevenly, so that a recurrent neural network model overfits most samples and underfits few samples, andinaccurate prediction is caused. According to the method, secondary sampling is carried out on the samples through resampling, so that the model is fully trained, and the prediction precision is effectively improved.