Datathon-TM-29

CTG Fetal Health Classification

This project predicts fetal health states (Normal, Suspect, Pathologic) from Cardiotocographic (CTG) data using machine learning. It was developed as part of Hackathon 2025.

---

Repository Contents

• ctg_hackathon.ipynb – main Jupyter notebook with the full pipeline
• CTG_clean.csv – cleaned dataset used for training
• metrics_table.csv / metrics_table.png – model performance results
• submission.csv – predictions in required format (id, NSP)
• README.md – project description

---

Approach

1. Data Cleaning: removed duplicates, constant features, and ID-like columns.
2. Preprocessing: filled missing values, standardized numeric features, one-hot encoded categorical features.
3. Splitting: 80/20 train–test split with stratified 5-fold cross-validation.
4. Models: compared Logistic Regression (baseline) and Random Forest (ensemble).
5. Evaluation: used Macro F1 (primary) and Balanced Accuracy (secondary) to handle class imbalance.

---

Results

• Random Forest was the best model.
• CV Macro F1 (5-fold): 0.9894
• CV Balanced Accuracy (5-fold): 0.9849
• Hold-out Macro F1: 0.9759
• Hold-out Balanced Accuracy: 0.9651

Confusion matrix shows high accuracy. Most errors were between Normal and Suspect, while Pathologic was identified well.

---

How to Run

1. Clone the repo and open in your environment.
2. (Optional) create a virtual environment:

   python -m venv .venv
   .venv\Scripts\activate  # Windows
   source .venv/bin/activate  # Linux/Mac