Construction of exchange rate forecasting combination model based on data mining technology

Due to frequent exchange rate changes, foreign exchange risks increase and foreign exchange speculation intensifies, which further aggravates the turmoil in the international financial market. RMB exchange rate is a complex nonlinear dynamic system, and its prediction belongs to time series. Aiming at the problems existing in the input data of time series data research, this paper will adopt the idea of fundamental analysis, select practical indicators to build an index system, and combine PCA (principal component analysis) and LSSVM (Least Squares Support Vector Machine) to form the structure of exchange rate forecasting combination model based on DM( Data Mining) technology, and establish PCALSSVM model. Bayesian evidence theory is used to maximize the probability of tuning parameters to obtain the optimal solution. The results show that the original data of PCA-LSSVM model has a good fitting degree with regression of 0.94842, which proves its good fitting degree. It is proved that it has small prediction error and high accuracy. The MAPE(Mean Absolute Percentage Error) value of PCA-LSSVM model is 0.00328, and that of SVM model is 0.00725. The MAPE value of GA-BPNN model is 0.00509. It can be seen that the prediction effect of PCA-LSSVM model is obviously better than SVM model. Through the analysis and comparison of examples, it is found that the combined forecasting model is more accurate than the single forecasting model, and it can predict the non-stationary exchange rate time series more accurately.