RonaldKato/linguistic-equity-maternal-health

Linguistic Equity and Cross-Lingual Risk Stratification in Multilingual Maternal Health Dialogues for Low-Resource African Languages # Linguistic Equity & Cross-Lingual Risk Stratification in Multilingual Maternal Health Dialogue Place `maternal_multilingual_dataset.py` (included) in the same folder, then run: pip install numpy pandas scikit-learn scipy networkx matplotlib seaborn --break-system-packages python3 main_pipeline.py This regenerates everything in `outputs/`: - `outputs/tables/*.csv` — 6 result tables (+2 supporting: compute cost, hyperparameters, Kruskal-Wallis) - `outputs/figures/*.png` — 12 analysis figures + 1 pipeline diagram - `outputs/computation_report.json` — wall-time, peak memory, and every hyperparameter used per stage ## Module map 1. data_loader.py — ingestion + repair of the raw dataset (3 real bugs fixed, documented in-file) 2. preprocessing.py — text cleaning, word + character n-gram TF-IDF feature spaces 3. cross_lingual_alignment.py — CL-SPA: Sinkhorn-refined Procrustes cross-lingual alignment (the core novel algorithm) 4. equity_metrics.py — Linguistic Equity Score (LES) + Kruskal-Wallis significance testing 5. topic_network_analysis.py — NMF topic modelling + Traditional-Practice Care-Seeking Exposure Network (TPCEN) 6. predictive_models.py — classifiers (Logistic Regression / Random Forest / Gradient Boosting) with GridSearchCV, + zero-shot cross-lingual transfer 7. evaluation_visualization.py — builds all tables and figures 8. main_pipeline.py — orchestrates 1-7 end to end, timed, with a computation report (run this file) Runtime on a single CPU core: ~3.2 minutes, peak ~1.2 GB RAM.