Diabetes Prediction AI — Complete Suite
Multi-Platform AI Prediction System | 95.2% Accuracy
Overview
Ensemble RF + XGBoost + LightGBM with SHAP; Discord bot, web UI, FastAPI, and Kivy mobile app with local SQLite for offline predictions.
Problem
Healthcare administrators need diabetes risk prediction across web, chat, API, and mobile platforms with explainable AI.
Solution
Ensemble model combines Random Forest, XGBoost, and LightGBM using 11 clinical features (age, BMI, HbA1c, cholesterol, etc.). Data validation enforces medical ranges. Discord bot accepts /predict commands, returns predictions with SHAP feature importance charts, and stores history. Web interface accepts manual forms or CSV batch uploads. FastAPI provides REST endpoints with API key authentication. Kivy mobile app stores predictions locally in SQLite for offline use.
Technologies
- Python
- scikit-learn 1.3
- XGBoost 1.7
- LightGBM 3.3
- SHAP 0.41
- FastAPI 0.104
- Discord.py 2.3
- Bootstrap 5
- Kivy 2.1
- Docker
Results
95.2% accuracy across 11 clinical features. Four interfaces: Discord bot, web dashboard, REST API, mobile app. SHAP explainability, batch CSV processing, offline mobile predictions.