This study by Miedema et al. investigates how supervised machine learning (ML) can classify mice by trait anxiety phenotypes with high accuracy, while simultaneously reducing the required sample sizes so as to better align with animal welfare principles. The authors pooled behavioral data from 470 mice across multiple cohorts who had undergone auditory aversive conditioning. They first used clustering to assign mice into anxiety phenotypes, then trained and validated sex-specific supervised ML classifiers such as logistic regression to ensure generalizability. The classifiers achieved high average accuracy, often greater than 90 percent in distinguishing sustained vs phasic responders, across both sexes and memory retrieval sessions. Furthermore, classifier performance remained accurate even when sample sizes were reduced, however clustering accuracy degraded with smaller cohorts. On independent validation datasets, the models matched clustering results with high consistency (Cohen’s κ > 0.885) and misclassified only a few animals. The authors highlight that applying ML can resolve latent behavioral phenotypes with greater reproducibility and lower animal numbers, helping align scientific rigor with the ethical principle of reduction. They argue that their research provides a practical and generalizable method to enhance phenotyping in clinical research and reduce reliance on large cohort size.