We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Insights into Performance Fitness and Error Metrics for Machine Learning

Abstract: Machine learning (ML) is the field of training machines to achieve high level of cognition and perform human-like analysis. Since ML is a data-driven approach, it seemingly fits into our daily lives and operations as well as complex and interdisciplinary fields. With the rise of commercial, open-source and user-catered ML tools, a key question often arises whenever ML is applied to explore a phenomenon or a scenario: what constitutes a good ML model? Keeping in mind that a proper answer to this question depends on a variety of factors, this work presumes that a good ML model is one that optimally performs and best describes the phenomenon on hand. From this perspective, identifying proper assessment metrics to evaluate performance of ML models is not only necessary but is also warranted. As such, this paper examines a number of the most commonly-used performance fitness and error metrics for regression and classification algorithms, with emphasis on engineering applications.
Comments: 18 pages, 2 tables
Subjects: Machine Learning (cs.LG); Data Analysis, Statistics and Probability (physics.data-an); Machine Learning (stat.ML)
Journal reference: 2021
DOI: 10.1007/s44150-021-00015-8
Cite as: arXiv:2006.00887 [cs.LG]
  (or arXiv:2006.00887v1 [cs.LG] for this version)

Submission history

From: M.Z. Naser [view email]
[v1] Sun, 17 May 2020 22:59:04 GMT (451kb)

Link back to: arXiv, form interface, contact.