Multiway Efficiency Evaluation Using Tensor-Based Data Envelopment Analysis

Traditional Data Envelopment Analysis (DEA) models are widely used for measuring the efficiency of decision-making units (DMUs), but their matrix-based structure limits their ability to analyze multidimensional data involving time, space, and contextual variations. To overcome this limitation, this study proposes a tensor-based DEA framework for multidimensional efficiency assessment. The proposed approach represents inputs and outputs as higher-order tensors and applies Tucker decomposition to capture latent relationships among variables, DMUS, and auxiliary dimensions. The framework introduces dynamic and context-aware efficiency weights that adapt across different conditions, enabling a more flexible and comprehensive evaluation compared with conventional DEA models. By preserving multidimensional interactions, the method can identify hidden efficiency patterns, temporal trends, and contextual effects that are not detectable using classical approaches. Experimental results on synthetic and real-world datasets demonstrate that the proposed framework achieves higher efficiency recovery accuracy and greater rank stability than Classical DEA, Dynamic DEA, and PCA-DEA methods. In addition, the model successfully distinguishes stable, improving, and declining DMUS over time while maintaining interpretability and computational scalability. The proposed tensor-based DEA framework provides an effective and interpretable solution for efficiency analysis in complex data-rich environments such as healthcare systems, financial institutions, and operational networks.

کلیدواژه ها:

Data Envelopment Analysis ، Tensor Decomposition ، Multidimensional Efficiency ، Tucker Decomposition ، Dynamic Efficiency Evaluation