Prediction of treatment effect perception in cosmetics using machine learning

Samir Salah*, Loic Colomb, Amelie Marie Benize, Celine Cornillon, Ayet Shaiek, John Charbit, Anna Schritz

*Corresponding author for this work

    Research output: Contribution to journalArticleResearchpeer-review

    2 Citations (Scopus)


    Perception of treatment effect (TE) in cosmetics is multifaceted and influenced by multiple parameters that need to be considered simultaneously when evaluating TE. Here we provide a global approach to predicting TE perception using Random Forest (RF) classifier. Data from three randomized double-blind clinical studies with a total of 50 subjects were used. Different products were applied to each facial cheek of subjects at each visit, and post-application photographs were taken. Nine primary endpoints relating to skin pores were assessed by a specific image analysis algorithm. Twenty judges evaluated the relative pore visibility in all possible pairs of cheek photographs. RF was used to construct a prediction model for TE perception based on the primary endpoints and judge’s evaluation. Intra-study product ranking was done using the Bradley-Terry model on mean judges’ predicted preference. RF demonstrated overall good accuracy in predicting TE perception. Applying RF technique not only addresses issues of multiplicity, nonlinearity and interactions between multiple criteria but also focuses decision-making on one discrete parameter thereby simplifying interpretability and allowing for more consumer-centered claim substantiation in clinical trials. Abbreviations: RF: Random Forest classifier; FDA: The US Food and Drug Agency; ID: Identifier; MCID: Minimal clinical important difference; Param: Parameter; PGIC: Patients’ Global Impression of Change; TE: Treatment effect; TRT: Treatment.

    Original languageEnglish
    Pages (from-to)55-62
    Number of pages8
    JournalJournal of Biopharmaceutical Statistics
    Issue number1
    Publication statusPublished - 2 Jan 2021


    • Cosmetics
    • claim substantiation
    • clinical evaluation
    • discrete choice modelling
    • machine learning
    • statistics
    • treatment effect perception


    Dive into the research topics of 'Prediction of treatment effect perception in cosmetics using machine learning'. Together they form a unique fingerprint.

    Cite this