README.md

1. Project Title

• EDEL (Error-Driven Ensemble Learning Algorithm)

2. Project Description

• EDEL is a machine learning algorithm designed to address the challenge of class imbalance. By dynamically reintroducing misclassified samples during training and employing a multi-perspective learning strategy, EDEL enhances the model's ability to recognize minority class instances. This project implements the EDEL algorithm and compares it with various baseline methods (such as SMOTE, SMOTE-TLNN-DEPSO, and models without any imbalance handling techniques) across multiple real-world datasets, including financial fraud detection and credit risk assessment.

3. Installation

• Step 1: Clone the repository:

  git clone <repository-url>

• Step 2: Set up the datasets. Refer to the Datasets download link.md file for download links to the datasets, and download them into the dataset folder under the respective directories.

4. Datasets

• Download datasets:
  • Open the Datasets download link.md file to get the download links for each dataset.
  • After downloading, place the datasets in their respective folders under the dataset/ directory (e.g., dataset/Adult Income Dataset, dataset/Spambase Dataset).
• Preprocessing the data:
  • For each dataset, run the preprocessing scripts located in the corresponding subfolder under the experiment folder to convert the raw data into data.csv.
  • Example command:

    python <dataset-folder>/preprocess.py

5. Usage

• Note: Before running the following scripts, make sure to update the dataset paths in baseline_models.py, experiment_main.py, experiment_smote.py, and experiment_smote_tlnn_depso.py according to the respective dataset locations.

• Baseline Model (No Imbalance Handling):

  • To run experiments using baseline models without any imbalance handling techniques:

    python experiment/baseline_models.py

• EDEL (Error-Driven Ensemble Learning):

  • To run experiments using the EDEL algorithm:

    python experiment/experiment_main.py

• SMOTE:

  • To run experiments using the SMOTE algorithm for imbalance handling:

    python experiment/experiment_smote.py

• SMOTE-TLNN-DEPSO:

  • To run experiments using the state-of-the-art SMOTE-TLNN-DEPSO method:

    python experiment/experiment_smote_tlnn_depso.py

6. Experiment Details

• Baseline Experiment:
  • baseline_models.py: Runs baseline models without any imbalance handling techniques across all datasets.
• EDEL Experiment:
  • experiment_main.py: Runs the EDEL algorithm on the datasets and records the results and metrics, such as Recall and G-mean.
• SMOTE Experiment:
  • experiment_smote.py: Runs experiments using the SMOTE algorithm for handling imbalance.
• SMOTE-TLNN-DEPSO Experiment:
  • experiment_smote_tlnn_depso.py: Runs experiments using the SMOTE-TLNN-DEPSO method, a state-of-the-art approach for handling class imbalance.

7. Sensitivity Analysis

• Sensitivity analysis for each dataset can be run using the scripts in the sensitivity_analysis/ folder. Each .py file corresponds to the sensitivity analysis for a specific dataset.
  • For example, to run the sensitivity analysis for the Adult Income dataset:

    python sensitivity_analysis/sensitivity_analysis_aid.py