<p>Traffic prediction plays a crucial role in optimizing transportation efficiency and urban mobility, serving as a core component of Intelligent Transportation Systems (ITS). Although existing methods have achieved substantial progress, most of them focus on time-domain modeling and lack effective utilization of global correlations in the frequency domain. To address the insufficiency in global modeling among frequency components, this paper proposes a novel traffic flow forecasting framework named Frequency Decomposition and Time–Frequency Feature Enhancement (FDTFE). The framework decomposes traffic data into low-frequency trends and high-frequency events using the Discrete Wavelet Transform (DWT), and introduces a Global Filter to enhance the global frequency features of each frequency component. For the trend and event components, a spatial-temporal attention mechanism and causal convolution are applied to further capture the long-term and short-term temporal dependencies, respectively. Finally, a cross-attention-based fusion module adaptively integrates the enhanced time–frequency features with the original input, while a Gated Temporal Unit (GTU) is used to refine the feature representations. Experimental results show that FDTFE outperforms existing prevailing methods on four public traffic datasets.</p>