Boosting Multi-Class Outpost Vector Generation (BMCOV) is proposed as a scalable framework for boundary-preserving multi-class classification and offers significantly improved computational efficiency. The method replaces the brute-force nearest neighbor search used in the original Multi-Class Outpost Vector Generation (MCOV) and its parallel variant (PMCOV) with a k-d tree-based strategy. This modification reduces the time complexity from \(O(N^2)\) to \(O(N \log N)\) and enables efficient processing of large and high-dimensional datasets. Experimental evaluations on eleven benchmark datasets indicate that BMCOV delivers substantially enhanced scalability and slightly improved classification accuracy compared to PMCOV. It achieves an average speedup of 1573 \(\times \) and 19 \(\times \) over the original PFMCOV and PAMCOV methods, respectively. BMCOV maintains a dataset size nearly identical to that of PMCOV. The framework offers tunable parameters for flexible control over boundary vector placement and supports adaptation to diverse data characteristics. These results demonstrate that BMCOV provides both scalability and boundary-preserving capability and that it is well suited for modern classification tasks involving complex or overlapping decision boundaries.
