Long-Term and Short-Term Prediction of VN-Index Using Machine Learning Models

Abstract

Long-term and short-term prediction of VN-Index helps investors make financial plans to manage risks. Normally, machine learning models agree on long-term and short-term prediction without analyzing the characteristics of different models that can forecast engineering features of different time series data cycles. The study evaluates and compares ARIMA, SVR, hybrid ARIMA-SVR, and LSTM algorithm models in the short and long term. Commonly, LSTM is known as a high-precision deep learning model for time series data, and ARIMA is used to process linear data linearly. The proposed models are analyzed and are used to predict VN-index data over short-time and long-term periods, including one day (short term), seven days and 30 days (long term). Based on the accuracy assessment metrics such as RMSE and MAE, the study concludes that LSTM is more suitable for long-term forecasting than ARIMA. In contrast, ARIMA has higher accuracy compared to LSTM in forecasting in the short term. The hybrid ARIMA-SVR model helps improve the expected performance of the ARIMA model because ARIMA handles the data linearity, while SVR handles the non-linearity part.