Malkia23/Spatial-AI-Cervical-Cancer-Botswana.

# Spatial-AI-Cervical-Cancer-Botswana. An implementation project from the Botswana–Harvard Partnership (BHP) Bioinformatics, AI and Data Science Training. Builds a transparent, geography-aware risk score to help the Ministry of Health and Wellness (MoHW) and BHP prioritise where to deploy mobile "see-and-treat" cervical cancer screening units across Botswana's districts. ## The problem Cervical cancer is Botswana's leading cause of cancer death among women, and over 70% of cases occur in women living with HIV. Botswana has only one radiation oncology referral centre, in Gaborone. That means women in remote districts (Kgalagadi, Ghanzi, North-West) can face a 5+ hour journey to reach it, contributing to late-stage diagnosis and worse outcomes. ## What this project does Combines three district-level factors into a single, explainable **risk score**: | Factor | Weight | Why it's included | |---|---|---| | Distance to the Gaborone referral centre | 45% | Direct proxy for screening/treatment access | | HIV prevalence | 35% | HIV suppresses HPV clearance; HPV causes nearly all cervical cancer, so HIV prevalence is a direct biological risk factor, not a general health indicator | | Population (inverse, as a rurality proxy) | 20% | Smaller, more dispersed populations tend to have thinner local health infrastructure | A **transparent weighted score** was used instead of a machine learning model because Botswana has only 10 administrative districts — an ML model trained on 10 rows would overfit and produce meaningless results. See Methodology for more. ## Repository contents ``` ├── Spatial_AI_Cervical_Cancer_Botswana.ipynb # Full analysis pipeline (Google Colab-ready) ├── district_risk_scores.csv # Output: ranked risk scores per district ├── risk_choropleth.png # Output: static risk map ├── interactive_risk_map.html # Output: interactive risk map ├── Cervical_Cancer_Screening_Policy_Brief.docx # One …