Corporate bankruptcy prediction based on the Adaboost algorithm for optimisation of long and short-term memory networks

Abstract

In this study, a Long Short-Term Memory (LSTM) network was used to model and predict time series data, providing an innovative approach to corporate bankruptcy prediction. Subsequently, the predictions and real labels of the LSTM models were used to train the Adaboost algorithm to improve the accuracy and robustness of the models. Eventually, multiple trained LSTM models are combined into a more robust integrated model by adjusting the weights. It is worth mentioning that the integrated model achieves 95.57% prediction accuracy on the training set and 94.39% prediction accuracy on the test set, which indicates that the model has good prediction effect and generalisation ability. The method proposed in this study is of great significance, firstly, by combining LSTM and Adaboost algorithm, we not only improve the accurate prediction ability of corporate bankruptcy, but also enhance the ability of identifying anomalies. Second, by combining multiple LSTM models and adjusting the weights to form a more powerful integrated model, we effectively improve the overall prediction performance. This approach can provide financial institutions, investors, and government regulators with a more reliable and accurate tool for assessing corporate bankruptcy risk, which can help identify potential risks and take appropriate measures in a timely manner.