This title slide is titled "Maternal Health Risk Classification." Its subtitle reads "Using Machine Learning to Predict Pregnancy Risk Levels."

Maternal Health Risk Classification

Using Machine Learning to Predict Pregnancy Risk Levels

Source: Team: Ahmed Al Nasser, Ali Alsadah, Mahdi Alhashem

The slide introduces the problem of maternal health risks during pregnancy and the goal of predicting risk levels using machine learning. It highlights the importance of early detection to save lives.

Introduction

• Problem: Maternal health risks during pregnancy.
• Goal: Predict risk levels using machine learning.
• Importance: Early detection saves lives.

The Dataset Overview slide presents a dataset with 1014 total samples, each featuring 6 health indicator inputs and classified into 3 risk levels (low, medium, high). It is divided into 5 cross-validation folds for model evaluation.

Dataset Overview

• 1014: Total Samples

Dataset instances

• 6: Input Features

Health indicators

• 3: Risk Classes

Low, medium, high

• 5: CV Folds

Cross-validation splits Source: Maternal Health Dataset

The slide introduces the K-Nearest Neighbors (KNN) classification algorithm, with an optimal K value of 5 identified via grid search. It reports high accuracy from 5-fold cross-validation and strong performance on the maternal health dataset.

KNN Model

• K-Nearest Neighbors (KNN) classification algorithm
• Optimal K value: 5 via grid search
• 5-fold cross-validation: high accuracy achieved
• Strong performance on maternal health dataset

The SVM model tested Linear, RBF (best), Poly, and Sigmoid kernels, achieving 69.95% accuracy with RBF. It used 5-fold cross-validation and demonstrated strong sensitivity/specificity across folds.

SVM Model

• Kernels tested: Linear, RBF (best), Poly, Sigmoid
• Achieved 69.95% accuracy with RBF kernel
• Conducted 5-fold cross-validation
• Strong sensitivity/specificity across folds

The MLP model uses two hidden layers (16 and 8 neurons with ReLU activation) and was trained via Adam optimizer, cross-entropy loss, and early stopping. It achieved 70.22% test accuracy with strong low-risk detection (moderate high-risk) in the confusion matrix, making it the best performer among tested models.

MLP Model

• Architecture: 2 hidden layers (16 & 8 neurons, ReLU)
• Training: Adam optimizer, cross-entropy loss, early stopping
• Test accuracy: 70.22%
• Confusion matrix: Strong low-risk detection; moderate high-risk
• Best performing model among tested

The "Comparative Analysis" slide compares model accuracies, with MLP achieving the highest at 70.22% and SVM at 69.95% using RBF kernel. It also identifies K=5 as the optimal neighbor count for KNN.

Comparative Analysis

• 70.22%: MLP Accuracy

Highest performing model

• 69.95%: SVM Accuracy

RBF kernel results

• K=5: Optimal KNN

Best neighbor count Source: Cross-Validation Results

The MLP model achieved the highest accuracy of 70.22%. Future work includes real-time monitoring, mobile app development, advanced models like CNN/LSTM, XAI, and cloud deployment to advance safer pregnancies through AI collaboration.

Conclusion & Future Work

**Key Findings

• MLP model achieved highest accuracy: 70.22%

Future Work

• Real-time monitoring system
• Mobile app development
• Additional features integration
• Advanced models (CNN/LSTM)
• Explainable AI (XAI)
• Cloud deployment

[Blue future icon]

Closing: Advancing safer pregnancies with AI.

Call to Action: Collaborate to revolutionize maternal health.**

Summarizing Impact & Next Steps

Source: Maternal Health Risk Classification: Using Machine Learning to Predict Pregnancy Risk Levels

This slide displays a prominent "Thank You!" title with a blue thank you icon. The subtitle invites "Any Questions?" and provides a contact placeholder.

Thank You!

Any Questions?

Contact: [placeholder]

[Blue thank you icon]

Source: Maternal Health Risk Classification Presentation - Closing Slide