A significant problem occurs with natural resources, such as air pollution caused by various environmental factors and climate change. Air pollution poses a major threat to human health and sustainability. The varying levels of air pollutants mix dynamically, increasing air pollution and impacting human health proportionally to their basic health conditions. For example, a severity level of the air pollution immediately affects an old person or someone with breathing issues and can lead to sudden death. To save people, it is essential to develop an accurate and timely forecasting system to mitigate its adverse effects and take immediate action. Conventional forecasting systems use statistical and basic AI methods, often struggle to process complex and large amounts of continuous data generated from the air. Also, spatiotemporal dependencies from the air quality data were not extracted. Thus, this paper proposed a hybrid DL model, integrating a CNN with LSTM to analyse and accurately forecast the severity levels of air pollution. Basically, CNN model helps to extracts the spatial features from the air quality data while the LSTM model used to extract the temporal dependencies. The proposed CNN-LSTM can provide a robust prediction model for air pollution. The CNN-LSTM model is evaluated by implementing it in Python and experimenting with real-world datasets from various surveillance monitoring stations. The overall performance of the proposed CNN-LSTM is compared with the standalone LSTM, CNN and traditional ML models such as RF and RIMA. The final result indicate that proposed DL-based hybrid CNN-LSTM model performs healthier than the others and obtains the highest forecasting accuracy.