Datasetpublic

mabera/nigeria-amr-classifier-v2-dataset

Nigeria AMR Classifier v2 Dataset Author: Hussein Adeiza (mabera) Role: Licensed Environmental Health Officer, Abuja Nigeria Built for: AutoScientist Challenge 2026, Part 2 — Science Category Dataset Description A closed-label antimicrobial resistance classification dataset combining two independently verifiable task types: individual MIC-based susceptibility classification (CLSI M100 breakpoints) and population-level resistance rate classification from real… See the full description on the dataset page: https://huggingface.co/datasets/mabera/nigeria-amr-classifier-v2-dataset.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

0likes96downloads

Dataset Card

Nigeria AMR Classifier v2 Dataset

Author: Hussein Adeiza (mabera) Role: Licensed Environmental Health Officer, Abuja Nigeria Built for: AutoScientist Challenge 2026, Part 2 — Science Category

Dataset Description

A closed-label antimicrobial resistance classification dataset combining two independently verifiable task types: individual MIC-based susceptibility classification (CLSI M100 breakpoints) and population-level resistance rate classification from real Nigerian national surveillance data (MAAP/Fleming Fund Country Report, 2022).

Unlike open-ended interpretation datasets, every classification here is deterministically checkable, not subjectively judged.

Files

amrcombineddataset.csv

The complete 34-row training file (prompt/completion/source), combining both task types.

amrbreakpointregistry.csv

The CLSI M100 (2025) breakpoint ground truth used for individual MIC classification.

amrnationalsurveillance.csv

28 real rows from Nigeria's national AMR surveillance report, covering 25 sentinel laboratories and 23,963 positive cultures (2016-2018). Every resistance rate independently recomputed from raw N/n counts and matched exactly against the report's own published percentages.

amrclinicalcontext.csv

11 real cited clinical outcome statistics used for external validation.

Verification

All 34 rows in this dataset were programmatically verified before training, demonstrated live in the accompanying Kaggle notebook: https://www.kaggle.com/code/yunusahusseinadeiza/notebookb703acf2e7

Data Quality Diagnostic

Run automatically via verify_full_dataset.py, included in this repo:

—Total rows: 34
—Duplicate prompts: 0
—Null/missing values: 0
—Label distribution: balanced across Susceptible/Intermediate/Resistant (MIC layer), and spread across the full 14.5%-81.6% resistance rate range (surveillance layer), avoiding a degenerate single-class dataset

Reproducibility

All build and verification scripts are included in this repository. Run python verify_full_dataset.py to independently re-derive every classification label from raw ground truth. See REPRODUCIBILITY.md for full pipeline documentation.

Key Cited Findings

—3rd-generation cephalosporin resistance in Enterobacterales: 67-73% (2016-2018)
—MRSA rates reached 81.6% by 2018
—Carbapenem resistance in Enterobacterales fell from 19.0% to 14.5% (2016-2018), one of the few improving trends
—Resistant infections carry an 84% higher mortality risk globally; 32.1% vs 18.8% 30-day mortality, MDR vs non-MDR bloodstream infections

Sources

—CLSI M100, 35th Edition (2025)
—MAAP/Fleming Fund Regional Grant (Round 1), Nigeria Country Report, 2022
—Poudel AN et al., PLoS One, 2023
—BMC Infectious Diseases, Bacterial-associated bloodstream infections in Lagos, Nigeria, 2025
—PMC12659808, Libya prospective cohort, 2022-2024