The rapid growth of economic and industrial activities in Jakarta both contribute to rising air pollution levels. Accurate air quality forecasting is necessary for effective pollution management. Traditional linear methods are often insufficient to handle the complexity of air pollution data. To address these challenges, machine learning methods have emerged as powerful alternatives. In this study, the hybrid convolutional neural network with long short-term memory (CNN-LSTM) was employed, with LSTM handling temporal dependency, and CNN is used to handle spatial dependency. This study aims to demonstrate the effectiveness of CNN-LSTM in capturing hidden patterns of air quality without additional variables to enhance its performance. Furthermore, this method has never been implemented in Jakarta. The model predicts five key air quality parameters, namely PM10, NO2, SO2, CO, and O3 using collected data from five air quality monitoring stations in the Jakarta, recorded daily from 2013 until 2021. Missing values were addressed using Kalman smoothing and the optimal model selection was based on the smallest RMSE and MAPE values. The CNN-LSTM slightly outperforms LSTM with more efficient computing time. Results show that historical data alone provides reliable forecast, with PM10 and O3 as the dominant parameters at the critical point. Using epoch = 100, forecasting shows that air quality in Jakarta tends to be healthy to moderate. This forecast highlights the efficiency of CNN-LSTM for air pollution forecasting and its potential support decision-making regarding air pollution management in Jakarta.