Cost-sensitive Boosting for Classification of Imbalanced Data

This work aims to reproduce the results of the cost-sensitive boosting approaches presented by the paper

Project Members

• Omar Arab Oghli
• Hussain Kanafani
• Muaid Mughrabi

How to Run

git clone https://gitlab.com/omar.araboghli/cost-sensitive-boosting.git
cd cost-sensitive-boosting
pip install -r requirements.txt
python src/app.py

Configure the App

The config.json file must be stored in the same directory where the app.py is.

{
    "app": {
        "cost_setup": [], // list of costs to loop over
        "ratio_setup": [], // list of imbalance ratios to loop over
        "ratios_cost_setup": [], // list of costs to loop over when app mode is imbalance_ratio
        "n_experiments": 10, // number of experiments to be done in order to compute the average of the computed measures
        "imbalance_ratio": false // if true, only the measures_vs_ratios will be computed and stored. Otherwise, measures_vs_costs and weights_vs_iterations will be computed and stored
    },
    "dataProcessor": {
        "dataDir": "path/to/datasets/dir",
        "testSetSize": 0.2, // 20% of the dataset will be splitted to test set
        "datasets": [ // list of dataset objects
            {
                "filename": "breast-cancer.data", // dataset file name
                // list of columns names that contains categorical feautures
                "categorical_features": ["age","menopause", "tumor-size","inv-nodes","node-caps","deg-malig"
                                        ,"breast","breast-quad","irradiat"]
            },
            // ...
        ]
    },
    "model": {
        "algorithms": ["adac1"], // list of possible algorithms [adaboost, adac1, adac2, adac3, adacost]
        "base_estimator": "DecisionTreeClassifier",
        "n_estimators": 10, // number of iterations that the algorithm has to do
        "learning_rate": 1.0
    }
}

References

• Paper
• Our source code based on this implementation
• Click here to take a look on our presentations, poster, and the paper we based on.