A Hybrid Model Using PCA and BP Neural Network for Time Series Prediction in Chinese Stock Market with TOPSIS Analysis

Table 14

Summary of different models.

Property	PCA_BP_LM BP_LM	PCA_BP_BR BP_BR	PCA_BP_SCG BP_SCG	Exponential smoothing ,
Advantage	Fast learning rate	The training requires less data and effectively solves the problem of data overfitting	Beneficial in large-scale problems, with a fast convergence rate and small iterative computation amount	The storage requirements of the data are low, and the calculation is simple. The only need to select the appropriate damping coefficient according to the characteristics of the data will have good prediction results without training
Disadvantage	High memory consumption	Require relatively longer training time than other algorithms	More generations are required in training	Do not consider the characteristic relationship between variables, but only approximations of mathematical expressions
Summary	Good prediction performance can be achieved without dimensionality reduction	The prediction accuracy is better with the PCA method	The data type, size, or parameter setting can cause poor prediction performance. Overall, this algorithm is not suitable for stock price prediction	The algorithm does not consider that the stock price is affected by multiple factors, and the prediction effect is excellent and convenient. Besides, the stock price prediction performs well when selecting an appropriate smoothing factor