Abstract:In the field of traffic flow prediction, data collection frequently suffers from missing data due to sensor malfunctions and external interference. Though existing data imputation methods have made some progress, problems still exist: the direct use of raw traffic flow data may lead to false and misleading correlations, global consistency is often neglected when inferring missing data, and low-quality imputed data can generate additional noise. To address these issues, this study proposes a low-rank and confidence-driven imputation (LRCDI) framework for traffic flow data. First, the low-rank feature extraction module is employed to eliminate redundant information from raw traffic flow data to extract core feature representations, achieving data dimensionality reduction and denoising. Second, to enhance the accuracy of missing data prediction, a Fourier sparsity-constrained loss module is introduced to more effectively capture the global structure of the data, reducing prediction biases caused by over-reliance on local information. Finally, a historical confidence-driven data imputation module is proposed to filter out low-quality imputed data, thereby preventing noise from compromising the accuracy of subsequent prediction tasks. Experimental comparisons between the proposed data imputation method and other advanced methods on multiple datasets demonstrate that the proposed method exhibits superior performance and can more effectively accomplish data imputation.