Predicting Future Trends with an Ensemble Approach in Time Series Forecasting

Abstract

Predicting economic and financial time series has consistently presented challenges due to their susceptibility to influences stemming from economic, political, and social variables. Consequently, individuals involved in speculative activities within financial markets often seek strong models. Many advanced algorithms exist for forecasting the behavior of financial markets. This article introduces an approach that combines adaptive boosting regression with linear regression as the foundational estimator. The objective is to employ this amalgamation technique to forecast future closing prices of NASDAQ, gold commodities, the British pound sterling relative to the US dollar, and the euro against the US dollar. Our methodology integrates seven technical indicators in training adaptive boosting regression to enhance precision in predicting future closing prices. The assessment incorporates four metrics— coefficient of determination, mean absolute percentage error, root mean squared error, and mean squared error—to compare different machine-learning models. The analysis of experiments indicates that our technique attains superior accuracy compared to adaptive gradient-boosting regression, extreme gradient-boosting, and linear regression. This research will enable investors to make well-informed choices regarding future closing prices for four datasets, enabling them to determine the optimal timing for entering or exiting the market.