Evaluation framework of hierarchical clustering methods for binary data

Using Simulated Pest Models and Biological Clustering Validation to Improve Zoning Methods in Site-Specific Pest Management

“…Despite this result, a general pattern was observed where partitioning essays intended to distinguish between pest presence and pest absence sites (i.e., binary, absence-only) were better resolved by hierarchical MC algorithms (i.e., pests 1, 3, 4, 5, and 6), whereas more sophisticated approaches (i.e., SOM, DIA, MCL) were necessary to find differentiated levels of pest presence within general presence clusters (i.e., pests 1, 4, 5 and 6). This trend is consistent with published research that highlights the efficiency of hierarchical MC algorithms (i.e., SL, CL, WL) to partition sets of well-separated binary data (whether biologically meaningful or not) [10,65,66], as well as that of partitioning and model-based algorithms (i.e., MCL, DIA, SOM) to perform this same task with data sets that include observations more closely positioned in statistical space [50,67,68]. It must be noted, however, that plenty of other research works successfully explore the implementation of hierarchical clustering methods to partition spatially close data sets, as well as partitioning and model-based approaches to partition well separated binary data sets.…”

Patterns of Dupuytren Disease in Fingers

“…First, errors in clustering methods for binary data may be quite substantial, although an evaluation revealed that our method, using the Jaccard distance and complete linkage, is one of the best methods to choose. 13 Second, this analysis will forcedly create clusters, even when in the data no natural clusters exist. 29 A third drawback is the inability to address risk factors with this explorative analysis.…”

“…11 To identify the possible occurrence of patterns in fingers with Dupuytren disease, a hierarchical cluster analysis was conducted, assuming that patterns would be similar in both hands. The measure of similarity between fingers was based on Jaccard, 12 and the complete-linkage method 13 was applied to form clusters of fingers. Agglomerative hierarchical clustering (from bottom to top) was used.…”

Learning Treatment Regimens from Electronic Medical Records

“…In this concrete work, we utilize the learned binary hidden vectors and select the hierarchical clustering algorithm to divide patients into groups. We use the Hamming distance as similarity measurement for binary vectors and the complete linkage which was reported to give low error rate for symmetric distance measurement [10].…”