Accurate forecasting of PM2.5 concentrations plays a pivotal role in mitigating health risks and optimizing air quality management strategies, especially in regions with severe pollution challenges. This study evaluates the performance of five deterministic and two probabilistic forecasting models, leveraging diverse datasets from Changsha and Seoul to account for varying urban air quality dynamics. BiLSTM and Transformer consistently performed the best of the deterministic models, achieving the lowest MAE, RMSE, and MAPE. In contrast, ELM showed the highest error rates, indicating its limitations in capturing the complexities of air quality data. In the probabilistic forecasting domain, QRNN outperformed BNN in Changsha by providing more accurate prediction intervals, while BNN demonstrated superior reliability in Seoul despite wider intervals. These results highlight the importance of model selection based on dataset characteristics and environmental context, emphasizing the need for both deterministic and probabilistic approaches to enhance the accuracy and adaptability of air quality forecasting. Ultimately, these improvements will support better decision-making in air quality management.