Model selection in linear regression using the MML criterion

Rohan A. Baxter, David L. Dowe

    Research output: Chapter in Book/Report/Conference proceedingConference PaperOtherpeer-review

    4 Citations (Scopus)


    We use the minimum message length (MML) criterion for model selection in regression problems. The MML criterion uses an accurate estimate of the length of a code for the parameters of a Bayesian density. We have applied the MML criterion to the classic problem of choosing the degree of a polynomial in lease-squares regression. The MML criterion performed well in Monte Carlo experiments with small and/or noisy data sets relative to the other criteria examined.

    Original languageEnglish
    Title of host publicationProceedings of the Data Compression Conference
    EditorsJames A. Storer, Martin Cohn
    PublisherIEEE, Institute of Electrical and Electronics Engineers
    Number of pages1
    ISBN (Print)0818656379
    Publication statusPublished - 1994
    EventData Compression Conference 1994 - Snowbird, United States of America
    Duration: 29 Mar 199431 Mar 1994
    Conference number: 4th (Proceedings)


    ConferenceData Compression Conference 1994
    Abbreviated titleDCC 1994
    Country/TerritoryUnited States of America
    Internet address

    Cite this