<p>Long-term time series forecasting plays a pivotal but challenging role in automated machine learning (AutoML) research. Forecasting correlated bivariate time series involves modeling and analyzing the dependencies between two historical time series. Temporal correlation, inter-series correlation and long-term dependency are among the major challenges in bivariate time series prediction. Currently, statistical learning-based forecasting models are primarily developed by professionals and requires significant manual effort, including model design, feature engineering, hyperparameter tuning, and model construction. The proposed AutoML approach is designed to assist non-professionals in achieving effective time series forecasts, particularly for applications in finance and economics. Within the AutoML framework, this study proposes the use of Long Short-Term Memory (LSTM) networks to forecast both short- and long-term-correlated bivariate time series, demonstrating superior performance compared with the traditional statistical methods, such as Vector Autoregressive (VAR) and Autoregressive Integrated Moving Average (ARIMA). To streamline AutoML process, the pre-specified hyperparameters are introduced to significantly reduce the need for extensive tuning. A comprehensive analysis of LSTM within AutoML framework is also conducted to evaluate its advantages and limitations. Extensive experiments conducted on both univariate and bivariate time series with varying degrees of correlation show that the proposed method effectively improves prediction accuracy and captures inter-series dependencies.</p>