Despite the growing knowledge and commitment to climate change, carbon dioxide (CO2) emissions continue to rise dramatically throughout the planet. In recent years, the consequences of climate change have become more catastrophic and have attracted widespread attention globally. CO2 emissions from the energy industry have lately been highlighted as one of the world's most pressing concerns for all countries. This paper examines the relationships between CO2 emissions, electrical energy consumption, and gross domestic product (GDP) in Bangladesh from 1972 to 2019 in the first section. In this purpose, we applied the fully modified ordinary least squares (FMOLS) approach. The findings indicate that CO2 emissions, electrical energy consumption, and GDP have a statistically significant long-term cointegrating relationship. Developing an accurate CO2 emissions forecasting model is crucial for tackling it safely. This leads to the second step, which involves formulating the multivariate time series CO2 emissions forecasting challenges considering its influential factors. Based on multivariate time series prediction, four deep learning algorithms are analyzed in this work, those are convolution neural network (CNN), CNN long short-term memory (CNN–LSTM), long short-term memory (LSTM), and dense neural network (DNN). The root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) are used to analyze and compare the performances of the predictive models. The prediction errors in MAPE of the CNN, CNN–LSTM, LSTM, and DNN are 15.043, 5.065, 5.377, and 3.678, respectively. After evaluating those deep learning models, a multivariate polynomial regression has also been employed to forecast CO2 emissions. It seems to have nearly similar accuracy as the LSTM model, having a MAPE of 5.541.