Repository logo
 
No Thumbnail Available
Publication

Contribution to the knowledge of hierarchical clustering algorithms and consensus clustering. Studies applied to personal recognition by hands biometrics

Use this identifier to reference this record.
Name:Description:Size:Format: 
TeseCompleta.pdf1.54 MBAdobe PDF Download

Abstract(s)

In exploratory data analysis, hierarchical clustering algorithms with its features can provide different clusterings when applied to the same data set. In the presence of several clusterings, each one identifying a specific data structure, consensus clustering provide a contribution to deal with this issue. The work reported here is composed by two parts: In the first part, we intend to explore the profile of base hierarchical clusterings, according to their variabilities, to obtain the consensus clustering. As a first result of our researches, we identified the consensus clustering technique as having better performance than the others, depending on the characteristics of hierarchical clusterings used as base. This result allows us to identify a sufficient condition for the existence of consensus clustering, as well as define a new strategy to evaluate the consensus clustering. It also leads to study a new property of hierarchical clustering algorithms. In the second part, we explore a real-world application. In a first analysis, we use data sets derived by biometrics extracted from hands for personal recognition. We show that the hierarchical clusterings obtained by SEP/COP algorithms, can provide results with great accuracy when applied to these data sets. Furthermore, we found an increased 100% of recognition rate, comparing to the ones found in literature. In a second analysis, we consider the application of consensus clustering techniques to the problem of the identification of people's parenting by the hands biometrics. The results obtained indicate that hand’s photography has information that allows the identification of people’s family members but, according to our data, we didn't have very positive results (we observed a probability of 95% of the parents, and 94% of a sibling to be in the half of the more similar hands) that we believe it’s due to the poor quality of the photographs we used. However, the results indicate that the technique has potential, and if the collection of photographs is made using a scanner with fixed pins, the hand may be an interesting alternative for the identification of parenting of missing children when it is applied the consensus clustering.

Description

Keywords

Clustering; Consensus; hand biometrics

Citation

Research Projects

Organizational Units

Journal Issue