Abstract

Multidimensional data cubes are essential components in data warehouses, enabling rich, OLAP-based analysis across dimensions such as time, location, and product category. However, the complexity that supports such analytical flexibility often leads to extreme sparsity—where the majority of cube cells remain empty or only partially filled. This sparsity can hinder the performance of downstream machine learning models, especially when valuable but infrequent patterns are lost during preprocessing. This paper introduces a neutrosophic-based framework for evaluating and managing sparse regions within OLAP cubes. Instead of treating all sparsity as noise, we propose a typology that distinguishes between three forms: semantic sparsity (expected and justifiable absences), non-informative sparsity (regions with little analytical value), and informative sparsity (sparse areas that still carry meaningful insights). Each substructure is modeled using neutrosophic logic, which assigns degrees of truth, indeterminacy, and falsity to reflect its analytical potential. A dedicated Neutrosophic Evaluation Algorithm is developed to classify each region using metrics such as semantic confidence, entropy, and a context-aware informativeness score. These metrics allow for nuanced decisions: preserving informative sparsity, eliminating irrelevant regions, and flagging ambiguous areas for further review. This approach shows how neutrosophic logic can offer a novel and effective way to handle sparsity in OLAP cubes, improving the relevance and robustness of machine learning pipelines trained on multidimensional data.

Keywords :

Neutrosophic sets , Machine learning , Data mining , Data warehousing , Artificial intelligence , Uncertainty modeling , Sparsity , Data cubes , Preprocessing , OLAP

Cite This Article As :

• MLA
• APA
• Chicago
• Harvard
• Vancouver
• IEEE