Pima Indians Diabetes Prediction – Azure Machine Learning

Architecture Overview

Table of Contents

• Summary
• Project Structure
• Requirements
• Dataset
• Pipeline Overview
• Model Deployment
• Testing the Endpoint
• Visualization (Power BI)
• Reproducibility Notice

Summary

This project uses the Pima Indians Diabetes Dataset to train and deploy a logistic regression model that predicts the onset of diabetes.
Two models were trained and compared:

• Logistic Regression (LR)
• Support Vector Machine (SVM)

The logistic regression model, having better performance, was deployed using:

• Azure Container Instances (ACI) for quick testing
• Azure ML Online Endpoints (v2) for production-grade deployment

---

Project Structure

File	Description
pimadiabetes.csv	Raw dataset from Kaggle
data_diabetesPima_clean.csv	Cleaned dataset ready for training
preprocessing.py	Python script for data cleaning and preparation
submit_preprocessing.ipynb	Notebook to run preprocessing as a job on Azure ML
train_and_log.py	Training script with MLflow logging
submit_train.ipynb	Notebook to submit training job to Azure ML
score.py	Script used for model inference (used by deployment endpoint)
requirements.txt	Pip dependencies for ACI deployment
environment.yaml	Conda environment for Online Endpoint (SDK v2)
endpoint.yaml	YAML file to define the Online Endpoint v2
deployment.yaml	YAML file to deploy the model to the endpoint
deploiement_aci_pima.ipynb	Notebook to deploy the model using ACI
test.json	JSON input file for testing the deployed model

---

Requirements

• Python ≥ 3.7
• Azure CLI installed and configured (az login)
• Azure ML CLI extension (az extension add -n ml)
• Active Azure subscription with an ML workspace

---

Dataset

Download the dataset from Kaggle:
https://www.kaggle.com/datasets/uciml/pima-indians-diabetes-database

Rename it to:

pimadiabetes.csv

Place it in the root directory of the project.

---

Pipeline Overview

1. Preprocessing

python preprocessing.py --input_path ./ --output_path ./

Or run on Azure ML:

submit_preprocessing.ipynb

Output:

• data_diabetesPima_clean.csv
• correlation_matrix.csv

---

2. Model Training

python train_and_log.py

Or run on Azure ML:

submit_train.ipynb

Output:

• metrics_summary.csv
• roc_combined.csv
• learning_curves_combined.csv

---

---

Model Deployment

A. Deployment to ACI

Run on Azure ML::

deploiement_aci_pima.ipynb

B. Production Deployment via Online Endpoint (SDK v2)

az login
az account set --subscription <YOUR_SUBSCRIPTION_ID>
az ml workspace set --name <YOUR_WORKSPACE_NAME> --resource-group <YOUR_RESOURCE_GROUP>

# Create the endpoint
az ml online-endpoint create --file endpoint.yaml --workspace-name <YOUR_WORKSPACE_NAME> --resource-group <YOUR_RESOURCE_GROUP>

# Deploy the model to the endpoint
az ml online-deployment create --file deployment.yaml   --workspace-name <YOUR_WORKSPACE_NAME>   --resource-group <YOUR_RESOURCE_GROUP>   --endpoint-name pima-diabetes-endpoint   --all-traffic

---

Testing the Endpoint

az ml online-endpoint invoke   --name pima-diabetes-endpoint   --request-file test.json   --resource-group <YOUR_RESOURCE_GROUP>   --workspace-name <YOUR_WORKSPACE_NAME>

Example test.json file:

{
  "data": [
    [6, 148, 72, 35, 0, 33.6, 0.627, 50]
  ]
}

---

Visualization (Power BI)

Use the following CSV files:

• data_diabetesPima_clean.csv
• correlation_matrix.csv
• metrics_summary.csv
• roc_combined.csv
• learning_curves_combined.csv

Power BI Dashboard Preview

Reproducibility Notice

To keep this repository lightweight:

Only the original dataset pimadiabetes.csv is included.
Intermediate files like:

• data_diabetesPima_clean.csv
• correlation_matrix.csv
• metrics_summary.csv
• roc_combined.csv
• learning_curves_combined.csv

... are generated by executing the scripts provided.

---

Author

Edoh Kodji
edoh.kodji@gmail.com