Nearest-neighbor guided evaluation of data reliability and its applications

Area of Science:

• Artificial Intelligence
• Data Science
• Machine Learning

Background:

• Ordered Weighted Averaging (OWA) operators increasingly incorporate data reliability.
• Existing data-oriented OWA operators like DOWA use centralized structures, neglecting local data consensus.
• Cluster-based OWA (Clus-DOWA) addresses this but has high computational costs.

Purpose of the Study:

• To propose a computationally efficient nearest-neighbor-based reliability measure for OWA operators.
• To generate OWA weights and decision-support explanations from this measure.
• To demonstrate the effectiveness of the proposed method in alias detection and unsupervised feature selection.

Main Methods:

• Developed a nearest-neighbor-based reliability assessment, avoiding costly clustering.
• Utilized a stress function to derive OWA weights and explanations.
• Applied the method to information aggregation for alias detection and unsupervised feature selection.

Main Results:

• The proposed nearest-neighbor reliability measure is more computationally efficient than clustering-based methods.
• The approach successfully generates OWA weights and decision-support explanations.
• Applied techniques demonstrated superior performance compared to conventional state-of-the-art methods in alias detection and feature selection.

Conclusions:

• The nearest-neighbor reliability measure offers an efficient alternative for OWA operators.
• This method effectively enhances data aggregation tasks like alias detection and unsupervised feature selection.
• The proposed techniques show significant potential for improving data analysis reliability and performance.