A-jcodes/WellaPath-Data-Science-for-AI-Driven-Digital-Health-Access-in-Underserved-Communities

Data science project where I worked on data pipelines, analytics, and AI-driven decision support for a digital health platform serving underserved populations in Africa. **Project Summary** This project simulates a real-world health-tech deployment scenario where insufficient medical data required the creation of a structured synthetic dataset to enable machine learning–based disease classification. I worked within the Data Science team supporting the development of an AI-assisted symptom triage system for WellaPath, a digital health platform serving underserved populations in Nigeria. The objective was to classify patient symptom profiles into 7 common diseases using supervised ML models and determine the most reliable algorithm for downstream deployment. **Link**: wellapath.org **Problem Context** Healthcare access in low-resource environments suffers from: 1. Lack of structured patient records 2. Delayed diagnosis 3. Symptom overlap across diseases 4. Data scarcity for ML training Because collected data was insufficient, I designed and used a synthetic data generation pipeline reflecting real Nigerian disease prevalence and symptom distributions to train robust models. Dataset used: Synthetic data generator design (attached) **Dataset Description** 1. 7 diseases: Malaria, Pneumonia, Typhoid Fever, Measles, Lassa Fever, Influenza, Diarrheal Disease 2. Binary symptom encoding (fever, cough, rash, vomiting, diarrhea, etc.) 3. Probabilistic symptom assignment per disease 4. Noise injection for realism (2%) 5. Balanced to reflect realistic but imperfect health data Each row represents one simulated patient. **Modeling Approach** Six classification models were trained and compared: | Model | Accuracy | Why It Matters | | ------------------- | -------- | ------------------------------------------------- | | Logistic Regression | **0.85** | Strong baseline, interpretable for medical logic | | SVM | 0.84 | Handles co …