A Selective Ensemble Learning Method for Time Series Forecasting

Time series forecasting, as an important field of data mining, is of great significance in revealing future trends. Traditional statistical models have limitations in handling nonlinear and non-stationary time series. In contrast, machine learning models, especially ensemble learning, have become a research hotspot due to their advantages in balancing biases of multiple models and improving overall performance. Aiming at the existing time series forecasting methods that have the problems of single model and insufficiently accurate forecasting results, this paper proposes a time series forecasting model based on selective ensemble learning (s-SMLE), which significantly improves the forecasting accuracy by combining the advantages of different component models through the steps of data preprocessing, model screening and model weight generation. The experiments verify the effectiveness of the proposed model, providing new ideas and methods for time series forecasting.