Attribute reduction for incomplete mixed data based on neighborhood information system

Abstract

In an era of data-based and information-centric Industry 4.0, extracting potential knowledge and valuable information from data is central to data mining tasks. Yet, the ambiguity, imprecision, incompleteness, and hybrid in real-world data pose tremendous challenges to critical information mining. Accordingly, we propose a new Max-Correlation Min-Redundant (MCMR) attribute reduction model from the uncertainty relation of attributes to avoid information loss in incomplete mixed data. Specifically, the neighbor relations are primarily developed based on the soft computing approach of the neighborhood information system, which divides the objects into neighborhood covers to maximize the utilization of the information in the incomplete mixed data. Then, we detailly analyze the internal and external consistency relationships of the four main uncertainty functions. Based on this, a new MCMR uncertain function is designed with maximum relevance and minimum redundancy. Experiments on nine real-world datasets validate the proposed model can improve data quality by mining critical information in classification tasks and achieving optimal performance with a minimum number of attributes.

Keywords:

• Attribute reduction
• incomplete mixed data
• neighborhood information system
• uncertainty function
• MCMR

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 https://archive.ics.uci.edu/

2 https://aistudio.baidu.com/aistudio/datasetdetail/40690

Additional information

Funding

This work is supported by Program of Song Shan Laboratory (Included in the management of Major Science and Technology Program of Henan Province [221100210700-2]).