Distance-based clustering challenges for unbiased benchmarking studies

Abstract

Benchmark datasets with predefined cluster structures and high-dimensional biomedical datasets outline the challenges of cluster analysis: clustering algorithms are limited in their clustering ability in the presence of clusters defining distance-based structures resulting in a biased clustering solution. Data sets might not have cluster structures. Clustering yields arbitrary labels and often depends on the trial, leading to varying results. Moreover, recent research indicated that all partition comparison measures can yield the same results for different clustering solutions. Consequently, algorithm selection and parameter optimization by unsupervised quality measures (QM) are always biased and misleading. Only if the predefined structures happen to meet the particular clustering criterion and QM, can the clusters be recovered. Results are presented based on 41 open-source algorithms which are particularly useful in biomedical scenarios. Furthermore, comparative analysis with mirrored density plots provides a significantly more detailed benchmark than that with the typically used box plots or violin plots.

Analysis of College Students’ Ability to Improve Innovation and Entrepreneurship Based on Constrained Clustering Algorithm

“…In 2010, the Ministry of Education put forward new requirements for it, requiring that more attention should be paid to it in colleges, and the education target has also changed from the original entrepreneurs to all college students. To the whole process of college students' talent training, at the same time, the core of it is to improve college students' awareness of it and it [1]. In 2016, the Ministry of Education once again focused on this, emphasizing the reform content of it, and should continuously enhance the spirit of independent innovation, entrepreneurial awareness and innovation, and entrepreneurship ability of students [2].…”

Identification of Explainable Structures in Data with a Human-in-the-Loop

“…Descriptions and access to typical density-and distance-based structures [31] and algorithms [32] is provided. In the subsequent work [33], the pitfalls and challenges of automated cluster detection or cluster analysis pipelines are highlighted. This work shows that…”

“…• Parameter optimization on datasets without distancebased structures, • Algorithm selection using unsupervised quality measures on biomedical data, and • Benchmarking detection algorithms with first-order statistics or box plots or a small number of repetitions of identical algorithm calls are biased and often not recommended [33]. This serves as a motivation to investigate HIL approaches for structure identification toward pattern recognition as opposed to automatic algorithmic detection [22].…”

Flow cytometry datasets consisting of peripheral blood and bone marrow samples for the evaluation of explainable artificial intelligence methods

“… Researchers can use cases of each sample to benchmark unsupervised machine learning methods (c.f. [9] ) because domain experts (i.e., clinicians) distinguish pB samples from BM samples, and leukemia BM samples versus non-leukemia BM samples based on distributions of biological cell populations by looking at two-dimensional scatter plots. As a consequence, clear, straightforward patterns in data that have a biological meaning are visible to the human eye (c.f.…”