AI Project: Evaluation of Diabetic Classification Models

This repository contains the code and analysis for an Artificial Intelligence course project (LM-66 Cybersecurity and Cloud Platforms) focused on evaluating machine learning models for diabetes classification.

Project Overview

The project utilizes the "Diabetes Dataset" to classify patients as diabetic or non-diabetic. The primary goal is to:

• Implement and compare Logistic Regression, Decision Tree, and Random Forest models.

• Evaluate performance using metrics such as accuracy, precision, recall, F1-score, and AUC-ROC.

• Perform hyperparameter tuning using RandomizedSearchCV.

• Analyze feature importance to identify key factors influencing diabetes prediction.

The implementation is in Python, leveraging libraries including scikit-learn, Pandas, NumPy, Seaborn, and Matplotlib.This work demonstrates the application of AI techniques to a healthcare problem, relevant to data analysis and predictive modeling.

Key Objectives

• Accuracy: To develop models that accurately classify patients with and without diabetes.

• Robustness: To ensure the models perform well under various conditions and with real-world data.

• Clinical Relevance: To minimize false negatives to avoid misdiagnosis of diabetic patients.

Dataset

The "Diabetes Dataset" from Kaggle was used.It contains diagnostic information for Pima Indian women aged 21 and older.

Dataset Characteristics

• 768 instances

• 9 columns (8 features, 1 target variable)

• Features include: Pregnancies, Glucose, BloodPressure, SkinThickness, Insulin, BMI, DiabetesPedigreeFunction, Age.

• Target variable: Outcome (diabetic or non-diabetic).

Workflow

1. Data Preprocessing:

   • Exploratory data analysis.

   • Outlier removal using the Interquartile Range (IQR) method.

   • Feature engineering, including age categorization.

   • Data normalization.

2. Dataset Splitting:

   • 75% training set, 25% test set. Okay, continuing the detailed README:

---

3. Model Implementation:

   • Logistic Regression: Implemented with L1 and L2 regularization.

   • Decision Tree: Tuned for optimal depth.

   • Random Forest: Ensemble of decision trees.

4. Hyperparameter Tuning:

   • RandomizedSearchCV was used to optimize hyperparameters for each model.

5. Model Evaluation:

   • Evaluation on the test set.

   • Metrics: Accuracy, Precision, Recall, F1-score, Confusion Matrix, AUC-ROC curve.

6. Feature Importance:

   • Analysis of feature importance for Random Forest and Logistic Regression to understand the influence of each variable on the prediction.

Models

Logistic Regression

• Description: A linear model for classification.

• Hyperparameters Tuned: Regularization type (L1, L2), C (inverse of regularization strength).

• Regularization: L1 and L2 regularization were applied to prevent overfitting.

Decision Tree

• Description: A tree-like model that makes decisions based on features.

• Hyperparameters Tuned: Max depth of the tree.

• Depth Control: Limiting the depth helps to avoid overfitting.

Random Forest

• Description: An ensemble of decision trees that improves generalization.

• Hyperparameters Tuned: Number of trees in the forest, max depth of trees, min samples split.

• Ensemble Method: Combining multiple trees reduces variance and improves robustness.

Results

For more details, you can consult the document “Evaluation of diabetic metrics” found in the Github repo

Conclusion

This project successfully implemented and evaluated machine learning models for diabetes classification. The results demonstrate the potential of AI in healthcare for early detection and diagnosis support.

Files

• main.py: Main script to run the analysis.

• DecisionTree.py: Contains the model implementation of Decision Tree.

• LogisticRegression.py: Contains the model implementation of Logistic Regression.

• RandomForest.py: Contains the model implementation of Random Forest.

• diabates.csv: File for the dataset.

Dependencies

• Python 3.12

• scikit-learn

• Pandas

• NumPy

• Matplotlib

• Seaborn

Author(s)

• Francesco Alfonso Barlotti

• Salvatore Colucci

• Domenico Falco

Acknowledgements

• Prof.ssa Loredana Caruccio

• Prof.ssa Genoveffa Tortora

• University of Salerno-Master's Degree in Computer Science LM-66 Cybersecurity and Cloud Platforms