In this work, we tested the capabilities of various Bayesian networks structures (mainly Naive Bayes and augmented Naive Bayes) in a classification task, over the standard Adult dataset, which aims at separating people whose income is greater than 50 thousands dollars per year from the rest.
In order to play with the provided Jupyter notebook and test the various classifiers, it is necessary to follow these steps:
- Install
Python 3.8on your system - Optionally create a virtual environment in the root directory of the project (
python3 -m venv venv) and activate it (source venv/bin/activate) - Install the required dependencies (
pip install -r requirements.txt)
- Naive Bayes (NB): Implementation given by
pgmpy - Tree-Augmented Naive Bayes (TAN): Implementation taken by a pending pull request on the
pgmpyrepository - BN-Augmented Naive Bayes (BAN): Custom implementation (slow and buggy)
- Forest-Augmented Naive Bayes (FAN): Custom implementation
The Adult dataset was downloaded from the UCI Machine Learning Repository and placed inside the dataset folder.
The whole project was written in the Jupyter notebook classify.ipynb, while the custom structural learning algorithms are implemented in the estimators.py file.
Moreover, a complete overview of the whole data pre-processing, classification and evaluation pipeline can be found in the report.pdf file, inside the report folder.