GDM Diagnosis Using Machine Learning

This repository implements and benchmarks multiple machine‑learning pipelines for early detection of Gestational Diabetes Mellitus (GDM). The work extends the empirical study published in Journal of Ayurveda and Integrative Medicine in 2024, where we demonstrated the usefulness of ML models alongside Ayurvedic management strategies for GDM.(pubmed.ncbi.nlm.nih.gov)

Repository structure

File	Purpose
pima-indians-diabetes.csv	Source dataset with eight clinical attributes and binary diabetes outcome
Imputations_with_RF_and_LR.ipynb	Baseline and model‑based imputation experiments
RandomForestIterativeImputer.ipynb	Iterative imputation with tree estimators
FeatureImportances.ipynb	Model‑driven feature importance analysis
Classifiers_with_wrapper.ipynb	Training and evaluation of individual classifiers using wrapper selection
Ensemble_Learning.ipynb	Bagging, boosting and stacking ensembles

Data

We use the Pima Indians Diabetes dataset (768 records, 8 predictors). Zeros in clinical columns except Pregnancies are treated as missing and replaced with median values before modelling.

Column	Description
Pregnancies	Number of pregnancies
Glucose	Plasma glucose concentration (2‑hour)
BloodPressure	Diastolic blood pressure (mm Hg)
SkinThickness	Triceps skin fold thickness (mm)
Insulin	2‑hour serum insulin (mu U/ml)
BMI	Body mass index (kg/m²)
DiabetesPedigreeFunction	Family history score
Age	Age in years
Outcome	Diabetes diagnosis label

Methods

Imputation

• Simple: mean, median and mode
• Model‑based: Random Forest and Logistic Regression regressors
• Iterative: chained Random Forest regressors

Models

• Logistic Regression
• Support Vector Machine
• K‑Nearest Neighbours
• Decision Tree and Random Forest
• Gradient Boosting (XGBoost or AdaBoost depending on environment)
• Voting and stacking ensembles

Metrics

• Accuracy
• Precision, Recall, F1
• ROC‑AUC
• Confusion matrix with five‑fold stratified cross‑validation

Quick results

Model	Accuracy	ROC‑AUC
Logistic Regression (scaled)	0.708	0.813
Random Forest (median impute)	0.740	0.816
Random Forest (raw zeros)	0.747	0.813

Top five features by Random Forest importance:

1. Glucose (0.27)
2. BMI (0.16)
3. Age (0.13)
4. DiabetesPedigreeFunction (0.13)
5. BloodPressure (0.09)

How to reproduce

git clone https://github.com/SnehaDharne/gdm‑diagnosis.git
cd gdm‑diagnosis

# create and activate environment
conda create -n gdmml python=3.9
conda activate gdmml
pip install -r requirements.txt

# launch notebooks
jupyter lab

Run notebooks in the sequence listed in the Repository structure section.

Citation

If you use this code or findings, please cite:

@article{shetty2024cdss,
  title={A machine learning‑based clinical decision support system for effective stratification of gestational diabetes mellitus and management through Ayurveda},
  author={Shetty, Nisha P and Shetty, Jayashree and Hegde, Veeraj and Dharne, Sneha D and Kv, Mamtha},
  journal={Journal of Ayurveda and Integrative Medicine},
  year={2024},
  volume={15},
  number={6},
  pages={101051},
  doi={10.1016/j.jaim.2024.101051}
}

Author

Sneha Dharne snehadattadharne@gmail.com • LinkedIn • GitHub

Feel free to open issues or pull requests to improve the project.