Machine Learning Project - Disease Detection with Genetic Algorithm Optimization

This project focuses on applying a genetic algorithm to optimize machine learning models for detecting three diseases:

1. Breast Cancer
2. Parkinson's Disease
3. PCOS (Polycystic Ovary Syndrome)

Datasets

The datasets used in this project are sourced from Kaggle:

• Breast Cancer Wisconsin Data: Features extracted from breast mass aspirates.
• Parkinson's Disease Detection: Vocal measurements from healthy and Parkinson's patients.
• PCOS Dataset: Clinical measurements for PCOS diagnosis.

Models and Genetic Algorithm Optimization

1. Breast Cancer Detection

Models Evaluated:

• Logistic Regression
• Random Forest
• AdaBoost
• Decision Tree
• K-Nearest Neighbors
• Support Vector Machine (Linear)
• Support Vector Machine (RBF)

The genetic algorithm is specifically applied to logistic regression, resulting in an accuracy improvement from 96.5% to 98.6%.

2. Parkinson's Disease Detection

Models Evaluated:

• Random Forest
• AdaBoost
• Gradient Boosting
• Decision Tree
• KNN
• Support Vector Machine (Linear)
• Support Vector Machine (RBF)

Genetic algorithm optimization, specifically with gradient boosting, improves accuracy from 89.8% to 93.9%.

3. PCOS Detection

Models Evaluated:

• Random Forest
• AdaBoost
• Gradient Boosting
• Logistic Regression
• Decision Tree
• Support Vector Machine (Linear)
• Support Vector Machine (RBF)
• KNN

Genetic algorithm optimization enhances KNN accuracy from 84.6% to 88.2%.

Genetic Algorithm Workflow

1. Initialization: Random population of binary chromosomes indicating selected/not selected features.
2. Fitness Calculation: Calculate fitness scores based on model accuracy.
3. Selection: Choose the best-scoring chromosomes.
4. Crossover: Create a new population through chromosome crossover.
5. Mutation: Introduce random mutations in chromosomes.
6. Repeat: Iteratively repeat steps 2-5 for multiple generations.

Results

Application of the genetic algorithm optimization provides notable improvements:

• Breast Cancer: Logistic Regression accuracy improves from 96.5% to 98.6%
• Parkinson's Disease: Gradient Boosting accuracy improves from 89.8% to 93.9%
• PCOS: KNN accuracy improves from 84.6% to 88.2%

The optimal feature subsets selected by the genetic algorithm are detailed in the notebook.

Usage

The IPython notebook contains the full code implementation. To use:

1. Install requirements: numpy, pandas, scikit-learn, matplotlib
2. Run Jupyter notebook
3. Run cells in order

The outputs include accuracy scores, confusion matrices, and feature importance graphs.

Conclusion

Using a genetic algorithm to optimize feature selection improves machine learning models for disease detection across different datasets. This demonstrates the effectiveness of the approach in healthcare applications.