P. De La Lama, J. A. Martin Fernandez, M. Comas Cufi
Compositional Data (CoDa) requires a specialized statistical framework to analyze data expressed as proportions. Unlike conventional methods, CoDa focuses on the relative relationships between components, which is crucial in geochemistry and genomics. The sample space of CoDa is the Simplex, necessitating methods that preserve data properties. Aitchison geometry provides a framework ensuring scale invariance and subcompositional coherence. Principal Balances (PBs) are linear combinations of logratios that capture variability while maintaining interpretability. The exhaustive method for calculating PBs guarantees optimal solutions but is computationally intensive for high-dimensional data. An alternative, the constrained method, improves efficiency but may not guarantee global optima. This work proposes a heuristic approach combining the constrained method with Tabu Search, achieving high-quality solutions in fewer iterations, and demonstrating efficiency for high-dimensional CoDa.
Keywords: Compositional Data, Principal Balances, Tabu Search
Scheduled
Non Parametric Statistics I
June 12, 2025 3:30 PM
Auditorio 1. Ricard Vinyes