Benchmarks
Rare earthsRare Earth Separation IntelligenceRecovery prediction

Rare Earth Recovery Prediction

Predict element recovery across a solvent-extraction separation train from stage conditions and feed assay. Lower MAE is better.

Metric

MAE (recovery %)

Baseline

Rare Earth Recovery Predictor

Submissions

37

Updated

2026-06-24

Leaderboard

RankSubmitterModelTypeScoreDate
1k.almeidasx-gbm-v3Research implementation1.422026-06-24
2h.novakstagenetResearch implementation1.552026-06-19
3materealree-recovery-predictorSynthetic-data model1.802026-06-18
4s.iyertabular-mlpBaseline model1.972026-06-10
5d.moreauridge-baselineBaseline model2.612026-06-02

Entries are reference and research submissions on the open split. Scores are illustrative and require plant-specific validation for industrial use.

Task definition

Evaluation metric

MAE (recovery %)

Submission rules

  1. 1

    Train only on the provided synthetic training split; the test split is held out.

  2. 2

    Report MAE on the official test split with a fixed random seed.

  3. 3

    Submit the model card, training data type, and license with each entry.

Discussion

  • s.

    s.iyer

    ML researcher · 2d

    Question

    Are ensemble submissions allowed, or single-model only? Worth stating explicitly in the rules.

    7
  • ma

    matereal

    Maintainer · 2d

    Comment

    Ensembles are allowed if reproducible from the submitted artifacts. We'll clarify in the next rules revision.

    5
Open by design. Private by principle.

Build on open metal intelligence, keep your plant advantage private

Explore open reference models and datasets, or start a private technical discussion. Open industry knowledge and private customer advantage are kept separate.