A scalable version of MADD for big-data classification
This work addresses the scalability bottleneck of traditional MADD (Median Absolute Deviation Distance) classifiers, whose computational complexity grows quadratically with sample size, rendering them impractical for high-dimensional large-scale data. To overcome this limitation, the authors propose a scalable MADD classifier that integrates representative sample selection with random Fourier feature approximation, enabling efficient application of MADD to massive datasets for the first time. The method preserves the strong discriminative power of MADD in high-dimensional spaces while substantially reducing computational overhead. Theoretical analysis and empirical evaluations demonstrate that the proposed approach achieves classification performance comparable to the original MADD with only a minimal increase in computational time, effectively balancing accuracy and efficiency.