Comparative Study on Time Series Deep Learning Based Financial Risk Early Warning Model Construction for High-Tech Enterprises

The short-term memory (LSTM) based time series deep learning early warning model is built in order to solve the heterogeneity and temporal transmission characteristics of financial risks in high-tech enterprises, which is the basis of what the paper puts together. Through the application of non-financial and multi-dimensional financial indicators, the attention mechanism of the dynamic risk development pattern is included. We then compare the results of XGBoost and logistic regression models with a comparison of the data. According to the results of the experiment, the LSTM model can achieve 97% recall and a 0.9908 AUC-ROC value on the test set. However, there are some important warning advantages. The study uses partial reliance plots and KernelSHAP to explain the interactive effects between indicators. These results give the technology companies with high-precision, easily interpreted, and dynamically decision support for the prevention of risk.