Fetal Health Classification

Overview

This project aims to classify fetal health using features extracted from Cardiotocogram (CTG) data. It categorizes fetal health status into three classes: Normal, Suspect, and Pathological. The goal is to enhance the accuracy and reliability of fetal health assessments, providing healthcare professionals with an effective tool for early diagnosis.

Table of Contents

1. Objectives
2. Data Source and Description
3. Data Preparation
4. Exploratory Data Analysis (EDA)
5. Machine Learning Models
6. Results
7. Usage
8. Documentation

Objectives

• Problem Objective: Monitor fetal health using CTG data.
• Research Questions:
  1. Can we accurately classify fetal health using CTG data?
  2. Which machine learning algorithms are most effective for this classification task?
  3. What are the most important features for predicting fetal health?

Data

Source

• The dataset is publicly available on Kaggle: Fetal Health Classification.

Description

• The dataset contains 2,126 records with features derived from CTG exams, which were then categorized into three classes by expert obstetricians: Normal, Suspect, and Pathological.

Features

• FHR baseline, accelerations, fetal movement, uterine contractions, light decelerations, severe decelerations, prolonged decelerations, abnormal short-term variability, histogram metrics, and others.

Target

• 'fetal_health' - Classified as 1 (Normal), 2 (Suspect), and 3 (Pathological).

Data Preparation

Preprocessing

• The dataset did not contain missing values. Standardization was applied to ensure consistency in feature scales.

Class Imbalance Handling

• Applied SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance, thereby enhancing the performance of models in predicting minority classes.

Exploratory Data Analysis (EDA)

Visualization

• Histograms and Box Plots: Used to understand feature distributions and identify significant outliers.
• Correlation Matrix: Visualized using a heatmap to understand feature relationships. Highly correlated features were dropped to avoid redundancy.

Key Insights

• Features like prolonged decelerations and abnormal short-term variability were found to be positively correlated with fetal health issues.

Machine Learning Models

Algorithms Implemented

• K-Nearest Neighbors (KNN)
• Gaussian Naive Bayes
• Random Forest
• Gradient Boosting
• Logistic Regression
• Linear Discriminant Analysis (LDA)
• Neural Network (MLP)
• Support Vector Machine (SVM)

Best Performing Models

• Random Forest: Achieved an accuracy of 94%, excelling across all classes.
• Gradient Boosting: Demonstrated strong performance, particularly in distinguishing between classes.
• Neural Networks: Provided balanced precision and recall, suitable for capturing complex non-linear relationships.

Results

Performance Metrics

• Classification reports and confusion matrices highlighted the strong performance of Random Forest and Gradient Boosting, especially in identifying "normal" and "pathological" cases.

Usage

To run the project:

1. Clone the repository:

   git clone https://github.com/YOUR_USERNAME/Fetal-Health-Classification.git

2. Navigate to the directory:

   cd Fetal-Health-Classification

3. Install the required dependencies:

   pip install -r requirements.txt

4. Run the Jupyter notebook for analysis:

   jupyter notebook Fetal_Health.ipynb

Documentation

For a more detailed analysis and discussion, please refer to the summary report: Summary Report (PDF)