A Short-term Time Series Predictive Algorithm Based on Rolling Prediction and PSO-SVR
Shaowei Pan, Bo Yang, Qian Song · 2024
To overcome the shortcomings of current methods, a new algorithm for short-term time series forecasting based on rolling prediction, support vector regression (SVR) and particle swarm optimization (PSO) is presented. In this algorithm, rolling prediction is applied to construct the dataset, SVR is used to construct the prediction model and PSO is used to help SVR determine the optimal values of the hyperparameters. This algorithm is applied to the prediction of the number of postgraduate applicants and postgraduate admissions in China each year, respectively, and good results are obtained. The RMSE, MAE and MAPE achieved by the prediction model that is based on this algorithm on the test dataset of postgraduate applicants and postgraduate admissions are 2.4435, 1.7136, 0.0223 and 1.7210, 0.9529, 0.0603, respectively, which are lower than those achieved by linear regression (LR), random forest (RF), SVR without hyperparameter optimization and SVR optimized only by the genetic algorithm (GA-SVR).