No abstract

A single prediction model has its own advantages and disadvantages in different aspects. In order to improve the accuracy of carbon trading price prediction, this paper proposes ARIMA-BP-SSALSTM dynamic weighted combination model that combines ARIMA model, BP model and LSTM model optimized by Sparrow Search Algorithm (SSALSTM). Based on information entropy theory, the greater the information entropy, the more the information provided and the greater the weight. This model uses information entropy to assign dynamic weight and weights to obtain the combined prediction value. The empirical results show that the ARIMA-BP-SSALSTM dynamic weighted combination model is more accurate than the ARIMA model, BP model and SSALSTM model.

In recent years, the scale of loans in China has been increasing, and so has the credit risk. Yet the credit risk assessment system is still in the initial stage of development. To improve the accuracy of default loan prediction, this paper proposes ISSA-XGBoost model based on an improved way of passing input parameters. The model uses the Sparrow Search Algorithm (SSA) to optimize the parameters of eXtreme Gradient Boosting (XGBoost). Since the search strategy of SSA includes convergence to the origin direction, this paper changes the form of input parameters. In the form, the closer the parameters in SSA are to the origin, the more XGBoost under these parameters can avoid overfitting. SSA searches the optimal with the changed parameter form and outputs feasible solutions. Then, the model transforms their solutions back to the original form and inputs them into XGBoost. After the above processes, SSA can avoid overfitting while searching for optimal solutions. Using the improved SSA to optimize XGBoost, the ISSA-XGBoost loan default prediction model is established. The empirical result shows that the model outperforms SSA-XGBoost under the four metrics: ACC, AUC, KS, and BS. And it is significantly better than that of XGBoost optimized by Particle Swarm Optimization (PSO-XGBoost). At the same time, compared with SSA-XGBoost, the AUC score difference between the training set and the test set of ISSA-XGBoost is smaller, which indicates that ISSA-XGBoost can better avoid overfitting.