A novel hybrid model to forecast the stock price based on CEEMDAN and support vector regression

Abstract

For the last few decades, predicting the financial time series such as stock prices remained an interesting area for researchers. Because of the nonstationary and nonlinear characteristics, it is difficult to predict its future trajectory accurately using simple time series or econometric models. Therefore, in this study an attempt has been made to forecast stock prices using an improved hybrid ensemble model based on data decomposition technique such as complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and well known supervised machine learning algorithm called support vector regression (SVR). To check the efficiency of the proposed model the KSE-100 index daily closing prices of Pakistan stock exchange (PSX) in the time interval January 1, 2019 to April 26, 2024 has been used. Comparison of the proposed hybrid CEEMDAN-SVR model is made with other models such as CEEMDAN-Decision Tree (DT), CEEMDAN-Random Forest (RF), CEEMDAN-K nearest neighbors (KNN), and CEEMDAN-Artificial Neural Network (ANN). It is evident from the empirical findings that the proposed model performs better in terms of accuracy metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The numerical values of these statistical metrics for our proposed CEEMDAN-SVR model are 1562.116, 1401.253, 2.489, and 0.976, which are the lowest compared to other hybrid models. Therefore, advised to the financial time series experts to predict the financial time series utilizing this novel hybrid model.