A novel dependency-oriented mixed-attribute data classification method

doi:10.1016/j.eswa.2022.116782

2022-08-01

PRE

Yulin He

Philippe Fournier‐Viger

Joshua Zhexue Huang *

Ponnuthurai Nagaratnam Suganthan

DOI：10.1016/j.eswa.2022.116782

原文链接

原文求助

摘要

En 中文

How to design an efficient method to handle mixed-attribute data classification (MADC) problems has becomea hot topic in data mining and machine learning. Current MADC methods mostly transform mixed-attributedata into discrete-attribute data or continuous-attribute data before classification algorithms are trained. Thediscretization of continuous-attribute data usually results in information loss, while the binarization of discrete-attribute data generally yield more discrete-attributes. To address these issues, this paper proposes a novelMADC method abbreviated as DO-RVFL-NBC, which is a Dependency-Oriented aggregation model of randomvector functional link (RVFL) network and naive Bayes classifier (NBC). First, the method transforms theoriginal mixed-attribute set into a dependent attribute set and an independent attribute set by consideringthe variation rates of dependence and independence, respectively. Second, a RVFL network is trained basedon the dependent attribute set where each attribute has a weight to represent its dependence importancedegree. Third, a weighted NBC is constructed by assigning the independence importance degrees as weightsfor the calculation of class-conditional probability. Finally, exhaustive experiments are conducted to validatethe feasibility, rationality, and effectiveness of the DO-RVFL-NBC method using 22 benchmark mixed-attributedata sets. Experimental results show that (1) dependence and independence exist in the original mixed-attributeset and can be effectively explored; (2) changes of attribute dependences can improve the generalizationcapabilities of the RVFL network and NBC; and (3) a statistical analysis indicates that DO-RVFL-NBC canobtain considerably better testing accuracies on the benchmark mixed-attribute data sets in comparison with13 other MADC methods. This demonstrates that DO-RVFL-NBC is a viable approach for MADC problems

Keyword:

Mixed-attribute data classification

Attribute independence

Random vector functional link network

Naive Bayes classifier

One-hot encoding

Attribute discretization