Bachelor's Thesis - Benchmarking the ASCSR+ Classifer System on a Variety of Public Datasets Against Other ML Approaches

This repository contains the complete codebase for my bachelor's thesis.

Repository Structure

The project is organized into three main directories:

1. Sci-kit Learn Pipeline (/sklearn_pipeline)

Contains the scikit-learn based pipeline used for the training and evaluating of machine learning models on multiclass classification tasks.

Models included:

• LogisticRegression
• DT
• NB
• RandomForest
• GP
• XGB
• ANN
• SVC
• XCS
• ExSTraCS
• HEROS

2. ASCSR+ Model (/CIDS)

CLI tool expanded version of the original implementaiton.

3. Image Feature Extraction (/img_preprocessing)

Tools and scripts for extracting features from image data using a ResNet101 or ResNet50.

Requirements

sklearn and image preprocessing:

matplotlib
numpy<2.0.0
optuna
sqlalchemy<2.0
plotly>=4.0.0
pandas>=1.5.2
pip
pycodestyle
scikit-learn>=1.1.3,<1.3.0
threadpoolctl==3.1.0
scipy>=1.8.0
seaborn>=0.11.0
skrebate==0.7
tqdm
wheel
pytest
xgboost
lightgbm
catboost
gplearn
ipython
fpdf
scikit-XCS
scikit-ExSTraCS
scikit-eLCS
kaleido
dask-jobqueue
dask
joblib
graphviz
bokeh
ipywidgets
skheros

ASCSR+:

• .NET 5.0 SDK or higher

Installation and Training

The subfolders each contain a README.md describing the individual installation and usage of each part

Additional Acknowledgements

This thesis also employed the STREAMLINE AUTOML pipeline for all binary classification tasks