Fetal Health Classification

Project Overview

This project aims to classify fetal health status based on Cardiotocography (CTG) features. By analyzing a dataset of fetal heart rate and uterine contraction patterns, the goal is to develop a machine learning model that can accurately predict fetal health as 'Normal', 'Suspect', or 'Pathological'. This is a critical task for obstetric monitoring and clinical decision-making.

---

Technical Highlights

• Dataset: Kaggle - Fetal Health Classification
• Size: 2126 entries, 22 columns.
• Key Features:
  • CTG Metrics: baseline value, accelerations, fetal_movement, uterine_contractions, and a host of histogram-based features.
• Approach:
  • Data Cleaning: The dataset was clean, with no missing values. The code drops duplicate rows.
  • Exploratory Data Analysis: A correlation heatmap, box plots, and pair plots were used to visualize data distributions and relationships between features and the target class. A count plot showed the class distribution, which is imbalanced.
  • Multi-class Classification: The target variable fetal_health has three categories: 'Normal' (1), 'Suspect' (2), and 'Pathological' (3). The code adjusts these to be zero-indexed (0, 1, 2) for model training.
  • Data Standardization: MinMaxScaler was applied to the features for normalization.
  • Model Used:
    • An XGBoost Classifier model was trained.
• Best Accuracy:
  • The XGBoost model achieved an accuracy of 96.0% on the test set. The model's performance demonstrates its strong ability to classify fetal health status.

---

Purpose and Applications

• Obstetric Monitoring: Provide a tool for clinicians to quickly and accurately assess fetal well-being from CTG data.
• Early Warning System: Assist in identifying at-risk fetuses, enabling timely medical intervention.
• Clinical Decision Support: Support data-driven decision-making in labor and delivery.
• Research: Serve as a foundational model for studying the relationship between CTG signals and fetal health outcomes.

---

Installation

Clone the repository and download the dataset.

Install the necessary libraries:

pip install pandas numpy seaborn matplotlib scikit-learn xgboost

---

Collaboration

We welcome contributions to improve the project. You can help by:

• Performing comprehensive hyperparameter tuning and cross-validation for the XGBoost model to ensure robustness.
• Exploring more advanced strategies for handling the class imbalance.
• Investigating the impact of different feature selection or transformation techniques on model performance.
• Adding explainability (e.g., SHAP or LIME) to understand which CTG parameters are the most critical for predicting fetal health status.