Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A comparison of machine learning methods for classification using simulation with multiple real data examples from

Mizanur Khondoker1, Richard Dobson2, Caroline Skirrow3

  • 1King's College London, Institute of Psychiatry, Department of Biostatistics, London, UK King's College London, Institute of Psychiatry, NIHR Biomedical Research Centre for Mental Health at the South London and Maudsley NHS Foundation Trust, London, UK mizanur.khondoker@kcl.ac.uk.

Statistical Methods in Medical Research
|September 20, 2013
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Adaptive Gaussian process search for simulation-based sample size estimation in clinical prediction models: validation of the pmsims R package.

BMC medical research methodology·2026
Same author

Pre-hospital neurological risk stratification at return of spontaneous circulation (ROSC) after out-of-hospital cardiac arrest - the Pre-MIRACLE<sub>2</sub> score.

Resuscitation·2026
Same author

Autistic and ADHD adolescents respond differently to upset in school: evidence from a new questionnaire co-designed from the neurodivergent point of view.

European child & adolescent psychiatry·2026
Same author

Predicting long-term poor outcomes in individuals at clinical high risk for psychosis using real-world clinical data: the OASIS1000 prospective study.

World psychiatry : official journal of the World Psychiatric Association (WPA)·2026
Same author

Inequalities in type 2 diabetes incidence in a multiethnic population: a cohort study investigating the impact of ethnicity, migration and mental health comorbidities.

Diabetologia·2026
Same author

Development and validation of a precision treatment rules for first-line antipsychotic recommendations in first episode psychosis jointly incorporating effectiveness, side effects and patient preferences.

Translational psychiatry·2026

This study compares machine learning algorithms like Random Forests (RF) and Support Vector Machines (SVM), finding SVM superior for larger datasets. Simulation studies offer a more objective comparison of algorithm performance.

Area of Science:

  • Computational biology
  • Bioinformatics
  • Machine learning

Background:

  • Comparative studies of machine learning algorithms often suffer from bias due to selective reporting and sampling errors.
  • Simulation studies are proposed as a more objective alternative for evaluating algorithm performance.

Purpose of the Study:

  • To objectively compare the classification performance of widely used machine learning algorithms: Random Forests (RF), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and k-Nearest Neighbour (kNN).
  • To evaluate algorithm performance under various conditions, including feature number, sample size, and data variability.

Main Methods:

  • Utilized massively parallel processing on high-performance supercomputers to compare generalization errors.
  • Investigated performance across diverse combinations of factors: number of features, training sample size, biological and experimental variation, effect size, replication, and feature correlation.
Keywords:
cross-validationelectroencephalogram (EEG)generalisation errormachine learningmagnetic resonance imaging (MRI)microarraystruncated distribution

Related Experiment Videos

Main Results:

  • Linear Discriminant Analysis (LDA) is optimal for smaller, correlated feature sets (features <= half sample size), offering stable error estimates.
  • Support Vector Machines (SVM with RBF kernel) significantly outperform LDA, RF, and kNN with larger feature sets (sample size >= 20).
  • kNN performance improves with more features, outperforming LDA and RF unless data variability is high or effect sizes are small. Random Forests (RF) showed marginal superiority over kNN in specific high-variability, small-effect-size scenarios.

Conclusions:

  • The choice of machine learning algorithm depends critically on dataset characteristics, particularly feature set size and sample size.
  • SVM demonstrates robust performance across a range of conditions, making it a strong candidate for many classification tasks.
  • Findings from simulation studies were validated using real-world datasets, reinforcing their applicability.