Predictive Modeling for Maternal Health Risk Assessment Using Wearable and IoT Data

Title: Predictive Modeling for Maternal Health Risk Assessment Using Wearable and IoT Data (Reproducible Code) Description (Zenodo):This upload contains a complete, reproducible Python codebase for maternal health risk prediction using wearable and IoT physiological time-series data. The repository implements an end-to-end machine learning workflow: synthetic data generation (HRV proxy, systolic/diastolic blood pressure, SpO₂, glucose), preprocessing and standardization, feature engineering for classical ML baselines, and sequence modeling using deep learning architectures. Implemented models Baseline feature-based models: Logistic Regression, Random Forest Sequence-based deep models: LSTM, BiLSTM Transformer Encoder for multivariate time-series classification Pipeline components Synthetic dataset generator with clinically plausible ranges and class-dependent shifts for maternal risk categories Train/validation/test stratified splits with fixed random seed for reproducibility Preprocessing (channel-wise standardization fitted on training set only) Feature extraction (summary statistics + trend features) for baseline models PyTorch training scripts for LSTM/BiLSTM/Transformer with model checkpointing Evaluation utilities with Accuracy, Macro-F1, Weighted-F1, ROC-AUC (OvR), classification report, and confusion matrix Plot utilities for model comparison (accuracy/F1/ROC-AUC bar charts) OutputsRunning the scripts produces trained model artifacts, metrics in JSON format, and publication-ready plots saved under the outputs/ directory. ReproducibilityAll experiments are controlled by a global seed and use a fixed train/validation/test split strategy. The synthetic data in this repository is intended for method benchmarking and open reproducibility; it is not clinical guidance and is not a substitute for validation on real-world patient datasets. How to run (minimal) Install dependencies: pip install -r requirements.txt Generate synthetic dataset: python -m src.generate_synthetic --out data/synthetic.pkl Train baselines: python -m src.train_baselines --data data/synthetic.pkl --outdir outputs/baselines Train deep models (example): python -m src.train_deep --data data/synthetic.pkl --model transformer --outdir outputs/transformer Create plots: python -m src.plots --results_glob "outputs/**/metrics.json" --outdir outputs/figures License: MITKeywords: maternal health, wearables, IoT, time-series, machine learning, LSTM, Transformer, predictive modeling Notes Data are synthetic and provided to enable full public reproducibility. Users working with real wearable/clinical data should follow applicable ethics, privacy, and regulatory requirements and conduct external validation.