Evaluation of Clustering Results and Novel Cluster Algorithms: A Metascientific Perspective
MCML Authors
Theresa Ullmann
Dr.
* Former Member
Abstract
Theresa Ullmann
Dr.
* Former Member
Abstract
This dissertation addresses the reliability of clustering results and the evaluation of new clustering algorithms, particularly in light of the replication crisis in scientific research. The first contribution presents a framework for validating clustering results using validation data, ensuring the replicability and generalizability of findings. The second contribution quantifies over-optimistic bias in microbiome research by analyzing the effects of multiple analysis strategies on unsupervised tasks, while the third contribution highlights the over-optimism in evaluating new clustering algorithms, using the example of the 'Rock' algorithm, and advocates for more rigorous and neutral benchmarking methods. (Shortened.)
BibTeXKey: Ull22