Binary Mexican Axolotl Optimization (BMAO)

A Bio-Inspired Algorithm for Feature Selection in Machine Learning

---

Overview

This repository contains the reference implementation of Binary Mexican Axolotl Optimization (BMAO), a metaheuristic designed for feature selection in supervised learning.

BMAO extends the Mexican Axolotl Optimization algorithm to a binary search space, modeling biological processes of the axolotl—reproduction, injury, and regeneration—to balance exploration and exploitation when searching for compact and predictive subsets of features.

The algorithm has been evaluated on a wide range of UCI-style datasets, showing improvements in classification accuracy and dimensionality reduction compared to using all features.

This work was presented at the 3rd International Conference on Machine Learning and Pattern Recognition (MLPR 2025, Kyoto, Japan) and published in the ACM ICPS proceedings.
📄 DOI: 10.1145/3760678.3760693

---

Repository Structure

Binary-MAO-main/
│── feature_selection_MAO.py    # BMAO pipeline (cross-validation + evaluation)
│── mao_binary.py               # Core binary MAO implementation
│── data_loader.py              # Dataset loader and preprocessing
│── requirements.txt            # Dependencies
│── all_datasets/               # Benchmark datasets (.dat, .csv, .xlsx)

Datasets are expected to have the class label in the last column.
Non-numeric labels are automatically mapped to integers for binary classification.

---

Installation

git clone https://github.com/<your-username>/Binary-MAO.git
cd Binary-MAO-main
pip install -r Binary-MAO-main/requirements.txt

Python ≥3.8 recommended.

---

Example Usage

The pipeline can be called directly from Python with any scikit-learn compatible classifier and metric:

import os, sys
from sklearn.svm import SVC
from sklearn.metrics import accuracy_score

# Add repo to path
sys.path.append("Binary-MAO-main")

from data_loader import process_files
from feature_selection_MAO import model_with_metaheuristic_feature_selection

# Load datasets
dfs, names = process_files("Binary-MAO-main/all_datasets")

# Select one dataset, e.g. Pima
df = dfs[names.index("pima")]

# Run BMAO
results = model_with_metaheuristic_feature_selection(
    datasets=[df],
    datasets_names=["pima"],
    model=lambda: SVC(kernel="rbf", gamma="scale"),
    mao_metric="alpha",                   # "alpha", "alpha-mean" or custom callable
    evaluation_metric=accuracy_score,     # any sklearn metric
    validation_method="stratified_kfold",
    validation_params={"n_splits": 5, "shuffle": True, "random_state": 42},
    pop_size=30,
    max_iter=50,
    early_stopping_steps=10
)

print(results["pima"])
# {'mean_metric': <cv_score>, 'selected_features': [..], 'processing_time': <seconds>}

The function returns a dictionary per dataset with:

• mean_metric → performance score (e.g. accuracy)
• selected_features → indices of chosen features
• processing_time → runtime in seconds

---

Citation

If you use this work, please cite:

@inproceedings{alarcon2025bmao,
  title={Binary Mexican Axolotl Optimization (BMAO): A Bio-Inspired Algorithm for Enhancing Machine Learning Performance through Feature Selection},
  author={Rodríguez, Noé Oswaldo and Rosas, Carolina and Alarcón, Antonio and Yáñez-Márquez, Cornelio and Villuendas-Rey, Yenny and Recio García, Juan Antonio},
  booktitle={Proceedings of the 3rd International Conference on Machine Learning and Pattern Recognition (MLPR 2025)},
  year={2025},
  doi={10.1145/3760678.3760693}
}

---

Authors

• Noé Oswaldo Rodríguez Rodríguez (CIC-IPN)
• Carolina Rosas Alatriste (CIC-IPN)
• Antonio Alarcón Paredes (CIC-IPN)
• Cornelio Yáñez Márquez (CIC-IPN)
• Yenny Villuendas-Rey (CIDETEC-IPN)
• Juan Antonio Recio García (UCM Madrid)

---

License

This project is licensed under CC BY 4.0.
See the Creative Commons Attribution 4.0 International License.