A multidimensional feature grouping sampling algorithm based on dynamic feedback of prior bias

Abstract

The rapid development of information technology has led to the generation of massive amounts of large-scale discrete-variable data. However, processing the entire dataset will consume a lot of computing resources and be computationally inefficient. Sampling techniques provide a cost-effective solution to reduce the computational complexity while maintaining the original properties of the data. In pursuit of efficiency and effectiveness, this article proposes a multidimensional feature grouping sampling algorithm based on dynamic feedback of prior bias (MFGS) for sampling discrete-variable data. The basic idea is dynamic feedback iterative sampling. To this end, we established a dynamic feedback correction mechanism based on prior bias, which can accurately locate the sampling feature channel of each iteration, calculate the sampling size of each subgroup, and achieve accurate and targeted cyclic optimization sampling. Meanwhile, MFGS is introduced with the idea of smoothing filtering, which removes redundant samples in the oversampling area and can accurately limit the overall sample size. In addition, we use the multidimensional Manhattan distance to establish a sampling bias evaluation index, which provides a calculation basis for feedback and correction. Finally, we designed three experiments to verify the effectiveness of the feedback correction mechanism and smoothing filtering, and evaluate the sampling accuracy, computational efficiency, and sampling accuracy of the method under additional constraints. The experimental results show that the dynamic feedback correction mechanism and smoothing filter are effective, and MFGS outperforms the compared state-of-the-art methods in terms of sampling accuracy, and its computational efficiency is significantly improved compared with clustering-based sampling methods.