Research on Data-Driven Prediction Algorithm Based on Machine Learning

The aim of this paper is to use data-driven algorithms for data analysis and development trend research using moving average models and hybrid models based on variational modal decomposition (VMD) and gated recurrent unit (GRU). First, by analyzing the daily K-line of the data, the 5-day, 10-day, and 20-day moving averages are extracted, and it is observed that the moving averages show an overall upward trend, and the 20-day moving average is relatively smooth, with a smaller variation. Subsequently, a combined VMD-GRU forecasting algorithm is proposed, which first utilizes the variational modal decomposition technique to decompose the historical daily moving average series into a number of relatively smooth sub-sequences, in order to reduce the interference of different trend information on the forecasting accuracy. Then, for each subsequence, a time series forecasting model is built using the GRU neural network, and the forecasting results of each subsequence are superimposed to obtain the overall forecast value. With the data from May 6 to May 28, the model was corrected for error indicators to improve the accuracy of the prediction. Ultimately, the corrected model is utilized to predict the data trends for the next 20 days. This study provides an empirical analysis of machine learning-based data-driven forecasting algorithms, demonstrating their effectiveness and application potential in data time series forecasting.