双聚类
计算机科学
数据挖掘
聚类分析
机器学习
流程图
集合(抽象数据类型)
子空间拓扑
概率逻辑
人工智能
相关聚类
CURE数据聚类算法
程序设计语言
作者
Marta D. M. Noronha,Rui Henriques,Sara C. Madeira,Luis Enrique Zárate
标识
DOI:10.1016/j.patcog.2022.108612
摘要
• We present a biclusters evaluation flowchart until their validation. • Survey of 7 major groups of measures and their impact on coherence and structure. • Less common evaluation measures, including statistical and theoretical, are covered. • No single measure can assess all bicluster types regardless of the data structure. • A structured view on the limitations arising in the development of algorithms. To understand how subspace clustering algorithms discover distinct bicluster types and how their effectiveness has been validated, we offer a systematic literature review on available merit functions and how they affect the biclustering task. The covered principles are structured within a methodology to show how evaluation and validation measures/metrics determine the bicluster coherence, ensuring the algorithm effectiveness, and the limitations reported in some selected works. The review did not find any metrics that can be used in a generic way to guarantee the effectiveness of a biclustering algorithm when compared to all others. Therefore, the choice of evaluation metrics must meet to specific objectives of the application. So in this work, we present the measures and metrics in 7 major classes, including metrics based on residues, score thresholding, plaid, and order-preserving constraints, space transforms, correlations, theoretical and probabilistic frames, and set operations.
科研通智能强力驱动
Strongly Powered by AbleSci AI