In this paper, the Adaboost algorithm is optimised to classify and predict the user's credit risk by combining the long and short term memory neural network LSTM. The dataset was firstly divided, transposed, normalised, tiled and format converted and then the model was trained and tested. During the training process, it is observed that the loss on the training set gradually decreases and the model gradually optimally fits the data and gradually converges to the optimal solution. The confusion matrix shows that the credit risk of 2914 customers is correctly predicted in the training set with an accuracy of 83.3%. The model performs well on the training set and is able to predict the credit risk of the customers accurately. On the test set, 1211 customers' credit risks were correctly predicted with 80.7% accuracy. Compared to the training set, the prediction on the test set has slightly decreased, but it still copes well with the demand of predicting customers' credit risk. This indicates that the model has some generalisation ability and can achieve better performance on unknown data. Overall, the Adaboost algorithm based on LSTM optimisation proposed in this paper shows high accuracy and reliability in the credit risk classification prediction task. By combining neural networks and traditional machine learning methods, it improves the model's ability to accurately predict the credit risk of customers, providing an effective solution in the financial field.