Exploiting Data Distribution: A Multi-Ranking Approach.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Data heterogeneity arises from diverse sources, locations, structures, and formats, complicating distributed data management.
• Effective distributed data management necessitates specialized integration and analysis techniques for coherent processing and a unified global view.
• Feature selection is crucial for optimizing data processing and model performance in distributed learning environments.

Purpose of the Study:

• To propose a novel research methodology for constructing multi-level attribute rankings specifically for distributed data.
• To address the challenges posed by data heterogeneity in distributed learning systems.
• To evaluate the effectiveness of the proposed attribute ranking methodology.

Main Methods:

• Data was dispersed using table division based on reducts from rough set theory.
• Local rankings were generated using machine learning models, specifically the greedy algorithm for decision rule induction.
• Gradient boosting and neural networks were employed as classifiers to validate the methodology.

Main Results:

• The research methodology successfully constructed multi-level attribute rankings for distributed data.
• Experiments demonstrated the effectiveness of local rankings in assessing data quality for global models.
• The proposed approach showed merit in handling data heterogeneity and improving distributed learning.

Conclusions:

• The developed methodology offers a robust solution for attribute ranking in heterogeneous distributed data environments.
• The findings highlight the importance of local data quality assessment for enhancing global model performance.
• This research contributes to more effective data processing and analysis in large-scale distributed systems.