Coffee Bag Premium Classifier
Binary classifier predicting whether a retail coffee bag is premium (1) or standard (0) from its brand, origin, roast, grind, and packaging facts.
Source data
Trained on ssg1/coffee-bags-tabular, a classmate-built
tabular dataset for 24-679 (Fall 2026), Carnegie Mellon University. is_premium is a fixed
threshold on price-per-ounce; the price columns that define it are excluded from the feature set
to avoid data leakage. Features: altitude_masl, weight_g, altitude_stated, brand, origin, processing, roast_level, grind.
Training
Selected via AutoGluon TabularPredictor AutoML search (presets="medium_quality",
eval_metric="balanced_accuracy", 5-minute time budget, bagging/stacking disabled). Best model:
LightGBM.
Test results (6 original bags)
| Metric | Value |
|---|---|
| Accuracy | 0.667 |
| Balanced accuracy | 0.667 |
| Precision (premium) | 0.667 |
| Recall (premium) | 0.667 |
| F1 (premium) | 0.667 |
Intended use and limitations
Classroom exercise (24-679). 356 rows total, a small, non-random sample of retail coffee bags. Not validated for real purchasing decisions.
Load
import cloudpickle, huggingface_hub
path = huggingface_hub.hf_hub_download("CXu0630/2026-24679-coffee-premium-classifier", "autogluon_predictor.pkl")
predictor = cloudpickle.load(open(path, "rb"))