Model card: TF-IDF logistic-regression baseline

Intended use

Research baseline for binary disaster-message classification and demonstration of reproducible, risk-aware inference.

Not intended for

Autonomous dispatch, medical triage, evacuation orders, policing, immigration enforcement or identifying individuals.

Training data

User-supplied CSV with text and binary target. No training dataset is bundled.

Model

Word and bigram TF-IDF features with class-balanced logistic regression.

Outputs

Probability of the disaster class and one of three decisions: disaster, not_disaster, review.

Evaluation

Accuracy, precision, recall, F1, macro F1, Brier score, ECE, coverage, selective accuracy and weighted error cost.

Limitations

Vocabulary drift, multilingual performance, sarcasm, event shift and intentional misinformation require separate evaluation.