🔀 Multiplex Classification Framework

The Multiplex Classification Framework is a novel approach to complex machine learning classification problems — designed for scenarios with a large number of classes and logical constraints among them, such as medical image classification.

📄 Published in Applied Ontology

---

🧠 The Problem

Standard classification approaches (binary, multiclass, multi-label) work well for simple problems. But real-world scenarios — especially in healthcare — often involve:

• A large number of classes
• Logical constraints among classes (mutual exclusivity, subsumption, co-occurrence)
• Severe class imbalances
• The need for confidence threshold selection in multi-label settings

The Multiplex Classification Framework addresses all of these challenges.

Main types of classification approaches and examples.

---

✨ Key Features

• Scalable — supports any number of classes and logical relations
• No confidence thresholds — eliminates threshold selection inherent in multi-label classification
• Class imbalance handling — innovative task-splitting approach for imbalanced datasets
• Modular — each submodel in the ensemble can be fine-tuned independently
• Data quality improvement — automatically removes incompatible label combinations
• Performance — significant improvements in scenarios with many classes and constraints

---

🔧 How It Works

1. Taxonomy Adaptation

The original taxonomy is restructured following the Multiplex framework, reflecting class hierarchies and logical constraints. An OWL file is created from this structure.

Black ovals represent classes; blue rectangles represent the basic classification tasks that compose the full problem.

2. Dataset Adaptation

The input dataset is transformed into a Multiplex dataset, with one column per classification model. Labels are added or removed based on logical constraints between classes.

3. Model Ensemble Training

Each model in the ensemble is trained independently, allowing tailored hyperparameter tuning per submodel. See the experiments/ folder for full training and inference examples.

---

🚀 Quick Start

1. Clone and install

git clone https://github.com/mauro-nievoff/Multiplex_Classification
pip install -r Multiplex_Classification/requirements.txt

2. Import

from Multiplex_Classification.multiplex import *

3. Adapt your dataset

mdp = MultiplexDatasetProcessor(
    input_owl_path='path_to_your_taxonomy.owl',
    input_csv_path='path_to_your_dataset.csv'
)

---

📁 Repository Contents

Item	Description
multiplex.py	Core framework classes
experiments/	Notebooks with experiments on HyperKvasir and MultiCaRe datasets
sample_owl_files/	Example input/output OWL files
requirements.txt	Dependencies

---

📄 Publication

If you use this framework, please cite:

Nievas Offidani, M. (2025). Multiplex Classification Framework: A Theoretical Approach
to Complex Classification Problems in Machine Learning.
Applied Ontology. https://doi.org/10.1177/15705838251340362

---

🔗 Related Work & Resources

• 🏥 MultiCaRe Dataset — applied for medical image classification with a 140+ class taxonomy

---

🤝 Contributing

Contributions, issues, and pull requests are welcome.

For questions or collaborations, reach out on LinkedIn.