Enhanced Dempster-Shafer Classifier with Metaheuristics Optimization Article Swipe

MAKE IMPACT

Flores Araya, Emilio , Ramos-Rojas, Loreto , Veloz Alejandro , Olivares Rodrigo · YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.5281/zenodo.17718303

Dempster-Shafer Classifier with Metaheuristic Optimization A high-performance, Scikit-learn compatible implementation of a classifier based on Dempster-Shafer Theory (DST). This model utilizes Metaheuristic Algorithms (RSA, HBA, OPA) to optimize belief mass assignments, minimizing Binary Cross-Entropy (BCE) loss. Key computational kernels are accelerated using Numba JIT compilation, ensuring competitive training and inference speeds comparable to standard machine learning models. 🚀 Key Features Hybrid Architecture: Combines Evidential Reasoning (DST) with Evolutionary Optimization. Scikit-learn API: Fully compatible with GridSearchCV, Pipeline, and other sklearn utilities. High Performance: Core mathematical operations (Dempster's combination rule, normalization) are JIT-compiled using numba. Multiple Optimizers: Includes three state-of-the-art metaheuristics: RSA: Reptile Search Algorithm (Default) HBA: Honey Badger Algorithm OPA: Orca Predation Algorithm Interpretable: The model learns explicit mass assignments for feature intervals, offering a "white-box" alternative to neural networks. 📦 Dependencies To use this classifier, you need the following libraries:   pip install numpy scikit-learn numba ⚙️ API Reference DempsterShaferClassifier(M=5, optimizer='RSA', n_iterations=100, n_agents=30, ...)   Parameter Type Default Description M int 5 Number of equidistant division points (bins) per feature. optimizer str 'RSA' The metaheuristic algorithm to use. Options: 'RSA', 'HBA', 'OPA'. n_iterations int 100 Maximum number of iterations for the optimization loop. n_agents int 30 Population size (number of agents) in the metaheuristic. random_state int None Seed for reproducibility. custom_dim int None (Advanced) Manually override search space dimension. ⚡ Quick Start Here is a minimal example of how to use the DempsterShaferClassifier with Iris Dataset. import numpy as npimport pandas as pdfrom sklearn.datasets import load_irisfrom sklearn.model_selection import train_test_split, GridSearchCV, StratifiedKFoldfrom sklearn.preprocessing import StandardScalerfrom sklearn.pipeline import Pipelinefrom sklearn.metrics import classification_report, accuracy_score # Import local class (Ensure ds_classifier.py is in the same directory)from ds_classifier import DempsterShaferClassifier def print_model_masses(pipeline, feature_names=None):    """    Prints the learned belief masses for each feature interval.        Extracts the optimized mass vectors from the fitted model and displays    them in a human-readable format to facilitate interpretation.    """    # Access the classifier step within the pipeline    model = pipeline.best_estimator_.named_steps['ds_clf']        if not model.is_fitted_:        return     # Retrieve model dimensions    n_feat = model.n_features_in_    n_class = len(model.classes_)    M = model.M        # Reshape the flat mass array: (Features, Classes + Omega, Intervals)    raw_masses = model._best_masses_    reshaped = raw_masses.reshape(n_feat, n_class + 1, M + 1)        # Transpose for easier iteration: (Features, Intervals, Classes + Omega)    final_struct = reshaped.transpose(0, 2, 1)     print("\n" + "="*60 + "\n🧠 INTERPRETATION: LEARNED BELIEF MASSES\n" + "="*60)        # Create labels for classes plus the Uncertainty (Omega) term    cls_labels = list(model.classes_) + ['Ω']        for f_idx in range(n_feat):        # Determine feature name        name = feature_names[f_idx] if feature_names is not None else f"F{f_idx}"        print(f"\n🔹 {name.upper()}")                # Get the calculated division points (bin edges) for this feature        divs = model._feature_division_points_[f_idx]                for b_idx in range(M + 1):            # Format the interval range string for display            if b_idx == 0:                rng_str = f"(-inf, {divs[0]:.2f}]"            elif b_idx == M:                rng_str = f"({divs[-1]:.2f}, +inf)"            else:                rng_str = f"({divs[b_idx-1]:.2f}, {divs[b_idx]:.2f}]"                        # Get masses for the current interval            m_vec = final_struct[f_idx, b_idx, :]                        # Format the mass values            m_str = " | ".join([f"{l}:{v:.2f}" for l, v in zip(cls_labels, m_vec)])&l

Type

other

Landing Page

https://doi.org/10.5281/zenodo.17718303

OA Status

green

OpenAlex ID

https://openalex.org/W7107865430

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W7107865430

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.5281/zenodo.17718303

Digital Object Identifier

Title

Enhanced Dempster-Shafer Classifier with Metaheuristics Optimization

Work title

Type

other

OpenAlex work type

Publication year

2025

Year of publication

Publication date

2025-11-26

Full publication date if available

Authors

Flores Araya, Emilio, Ramos-Rojas, Loreto, Veloz Alejandro, Olivares Rodrigo

List of authors in order

Landing page

https://doi.org/10.5281/zenodo.17718303

Publisher landing page

Open access

Yes

Whether a free full text is available

OA status

green

Open access status per OpenAlex

OA URL

https://doi.org/10.5281/zenodo.17718303

Direct OA link when available

Concepts

Metaheuristic, Artificial intelligence, Classifier (UML), Computer science, Machine learning, Inference, Simulated annealing, Equidistant, Pattern recognition (psychology), Algorithm, Mathematical optimization, Artificial neural network, Parallel metaheuristic, Binary number, Feature (linguistics), Key (lock), Data mining, Training set, Pareto principle, Optimization problem, Binary classification, Feature selection, Automation, Evolutionary algorithm, Supervised learning, Global optimization

Top concepts (fields/topics) attached by OpenAlex

Cited by

0

Total citation count in OpenAlex

Full payload