Project

This is the final project for the Multiple Classifiers System's class.

Description

The goal of this project is to implement the proposed method of the paper "A two-stage ensemble method for the detection of class-label noise" and reproduce the experiments in some of the paper datasets, which can be found in the UCI Machine Learning repository. The experimented datasets were: Blood, Breast, Chess, Heart, Ionosfere, Liver, Parkinsons, Sonar and Spambase.

Getting Started

Requirements

• Python >= 3.7.1
• NumPy >= 1.15.4
• SciPy >= 1.1.0
• pandas >= 0.23.4
• scikit-learn >= 0.20.1
• matplotlib >= 3.0.1

Installing

• Clone this repository into your machine
• Download and install all the requirements listed above in the given order
• Download the listed datasets in .txt format
• Place all these files inside the data/ folder
• Change their file types to .csv
• Change their filenames according to the names in the ConfigHelper function get_datasets

Reproducing

• Enter into the project main folder in your local repository
• Run this first command to generate all metrics

python main.py

• Run this second command to aggregate these metrics and generate the final error table and graphics

python aggregate.py

Project Structure

.            
├── data                                  # Datasets files
├── results                               # Results files
├── src                                   # Source code files
|   ├── aggregate.py
|   ├── main.py
|   ├── majority_filtering.py
|   ├── noise_detection_ensemble.py
|   ├── config_helper.py
|   ├── data_helper.py
|   ├── io_helper.py
|   └── metrics_helper.py 
├── LICENSE.md
└── README.md

Author

• jpedrocm

License

This project is licensed under the MIT License - see the LICENSE.md file for details.