Sandbox Protocol_Active

Diabetes XAI API / BATCH_005

Protocol_Brief

Challenge

A simple 'Yes/No' isn't enough for medical AI. I wanted to provide 'Why' for every prediction.

Solution

Integrated SHAP explainability directly into the API response to highlight top physiological risk factors.

“I built this to learn how to make AI predictions transparent, which is critical in healthcare. I explored SHAP (Explainable AI) to show exactly why a model predicts a certain risk level. This project taught me about class imbalance handling with SMOTE and how to use Optuna for Bayesian hyperparameter optimization to reach high recall.”

Learned to use SHAP values to provide human-readable reasoning for ML predictions.
Mastered hyperparameter tuning using Bayesian optimization with Optuna.
Handled extreme class imbalance using SMOTE and KNNImputation techniques.
Built a production-ready API with FastAPI and containerized it with Docker.

Key_Modules

ReasoningAI LogicAPI

Loadout_Specs

PythonFastAPIXGBoostOptunaSHAPDocker

Sys_Diagnostics

Node_Count:5

Fidelity:Explainable AI (XAI)

Matrix_State:STABLE

db-pred.ais-cloud.dev

Viewport: Matrix_Simulated

Step 01

Optuna Tuning

See how I used automated tuning to maximize the model's diagnostic recall.