Metaheuristic Algorithm Based Feature Selection for Improving the Performance of Machine Learning Algorithms

Sinan Balıbey; Feyza Altunbey Özbay; Erkan Tanyıldızı

Makale detayları

Metaheuristic Algorithm Based Feature Selection for Improving the Performance of Machine Learning Algorithms

Yazarlar:

Sinan Balıbey | Feyza Altunbey Özbay | Erkan Tanyıldızı

Yayıncı:

Fırat University

Ana Başlıklar:

Yapay Zeka (Diğer)

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.509-522

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001334889

pISSN:

1308-9072

Anahtar kelimeler:

Özet:

 In a world of increasing data, the success of machine learning ML algorithms depends heavily on the process of feature selection. Feature selection aims to create a smaller and more meaningful feature set by eliminating redundant, unlabeled, or noisy features. The aim of this study is to obtain the most meaningful features from data sets with metaheuristic algorithms and to ensure that ML algorithms achieve better performance according to different evaluation metrics. In this context, metaheuristic algorithms for feature selection, Binary Particle Swarm Optimization, and Genetic Algorithms are used for experimental research on two different datasets. In the Sonar and Breast Cancer datasets, K-Nearest Neighbor, Support Vector Machine, and Decision Tree classification algorithms were used as the objective function. For both algorithms, feature selection was performed at high iterations with different population sizes, and then the best-selected features were evaluated with different ML models Logistic Regression, Naive Bayes, Extreme Gradient Boosting, Artificial Neural Networks, Convolutional Neural Networks, and Long Short Term Memory. The results show that Binary Particle Swarm Optimization and Genetic Algorithm methods lead to significant increases in model accuracy after feature selection. The results obtained are aimed at contributing to researchers using metaheuristic algorithms to improve the feature selection process. These findings support the effectiveness of metaheuristics in the feature selection process and the potential for performance improvement when combined with various ML models.

 In a world of increasing data, the success of machine learning ML algorithms depends heavily on the process of feature selection. Feature selection aims to create a smaller and more meaningful feature set by eliminating redundant, unlabeled, or noisy features. The aim of this study is to obtain the most meaningful features from data sets with metaheuristic algorithms and to ensure that ML algorithms achieve better performance according to different evaluation metrics. In this context, metaheuristic algorithms for feature selection, Binary Particle Swarm Optimization, and Genetic Algorithms are used for experimental research on two different datasets. In the Sonar and Breast Cancer datasets, K-Nearest Neighbor, Support Vector Machine, and Decision Tree classification algorithms were used as the objective function. For both algorithms, feature selection was performed at high iterations with different population sizes, and then the best-selected features were evaluated with different ML models Logistic Regression, Naive Bayes, Extreme Gradient Boosting, Artificial Neural Networks, Convolutional Neural Networks, and Long Short Term Memory. The results show that Binary Particle Swarm Optimization and Genetic Algorithm methods lead to significant increases in model accuracy after feature selection. The results obtained are aimed at contributing to researchers using metaheuristic algorithms to improve the feature selection process. These findings support the effectiveness of metaheuristics in the feature selection process and the potential for performance improvement when combined with various ML models.

 In a world of increasing data, the success of machine learning ML algorithms depends heavily on the process of feature selection. Feature selection aims to create a smaller and more meaningful feature set by eliminating redundant, unlabeled, or noisy features. The aim of this study is to obtain the most meaningful features from data sets with metaheuristic algorithms and to ensure that ML algorithms achieve better performance according to different evaluation metrics. In this context, metaheuristic algorithms for feature selection, Binary Particle Swarm Optimization, and Genetic Algorithms are used for experimental research on two different datasets. In the Sonar and Breast Cancer datasets, K-Nearest Neighbor, Support Vector Machine, and Decision Tree classification algorithms were used as the objective function. For both algorithms, feature selection was performed at high iterations with different population sizes, and then the best-selected features were evaluated with different ML models Logistic Regression, Naive Bayes, Extreme Gradient Boosting, Artificial Neural Networks, Convolutional Neural Networks, and Long Short Term Memory. The results show that Binary Particle Swarm Optimization and Genetic Algorithm methods lead to significant increases in model accuracy after feature selection. The results obtained are aimed at contributing to researchers using metaheuristic algorithms to improve the feature selection process. These findings support the effectiveness of metaheuristics in the feature selection process and the potential for performance improvement when combined with various ML models.

 Artan veri dünyasında, makine öğrenmesi ML algoritmalarının başarısı büyük ölçüde öznitelik seçimi süreçlerine bağlıdır. Öznitelik seçimi, gereksiz, etiketsiz veya gürültülü öznitelikleri eleyerek daha küçük ve anlamlı bir öznitelik kümesi oluşturmayı hedefler. Bu çalışmanın amacı, metasezgisel algoritmalar sayesinde veri kümelerinden en anlamlı öznitelikleri elde ederek ML algoritmalarının farklı kıyaslama ölçütlerine göre daha iyi performans elde etmesini sağlamaktır. Bu kapsamda, öznitelik seçimi için metasezgisel algoritmalardan İkili Parçacık Sürü Optimizasyonu ve Genetik Algoritma kullanılarak iki farklı veri seti üzerinde deneysel araştırmalar gerçekleştirilmiştir. Sonar ve Meme Kanseri veri setlerinde, K-En Yakın Komşu, Destek Vektör Makinesi ve Karar Ağaçları sınıflandırma algoritmaları amaç fonksiyonu olarak kullanılmıştır. Her iki algoritma için de farklı popülasyon büyüklüğünde yüksek iterasyonlarda öznitelik seçimi yapılmış, ardından seçilen en iyi öznitelikler farklı ML modeli Lojistik Regresyon, Naive Bayes, Aşırı Gradyan Arttırma, Yapay Sinir Ağları, Evrişimsel Sinir Ağları ve Uzun Kısa Süreli Bellek ile değerlendirilmiştir. Sonuçlar, İkili Parçacık Sürü Optimizasyon ve Genetik Algoritma yöntemlerinin öznitelik seçimi sonrası model doğruluklarında anlamlı artışlara yol açtığını göstermiştir. Elde edilen sonuçların metasezgisel algoritmalar kullanarak özellik seçim sürecini geliştiren araştırmacılara katkı sağlaması amaçlanmaktadır. Bu bulgular, metasezgisel algoritmaların öznitelik seçimi sürecinde etkinliğini ve çeşitli ML modelleri ile birleşmesinin performans iyileştirmedeki potansiyelini desteklemektedir.

Makale detayları

Metaheuristic Algorithm Based Feature Selection for Improving the Performance of Machine Learning Algorithms

Yazarlar:

Sinan Balıbey | Feyza Altunbey Özbay | Erkan Tanyıldızı

Yayıncı:

Fırat University

Ana Başlıklar:

Yapay Zeka (Diğer)

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.509-522

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001334889

pISSN:

1308-9072

Metaheuristic algorithms

Feature selection

Machine learning algorithmsMetasezgisel algoritmalar

Öznitelik seçimi

Makine öğrenmesi algoritmaları

Sayı

Makale detayları

Metaheuristic Algorithm Based Feature Selection for Improving the Performance of Machine Learning Algorithms

Yazarlar: Sinan Balıbey | Feyza Altunbey Özbay | Erkan Tanyıldızı

Yayıncı: Fırat University

Ana Başlıklar: Yapay Zeka (Diğer)

Dergi: Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.509-522

Belge Tipi: Makale

Yayın Tarihi: 3/2025

Belge Dili: Türkçe

Document ID (Org No.): 2001334889

pISSN: 1308-9072

Metaheuristic algorithms

Feature selection

Machine learning algorithmsMetasezgisel algoritmalar

Öznitelik seçimi

Makine öğrenmesi algoritmaları

Sayı

Yazarlar:

Sinan Balıbey | Feyza Altunbey Özbay | Erkan Tanyıldızı

Yayıncı:

Fırat University

Ana Başlıklar:

Yapay Zeka (Diğer)

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.37 Issue 1, pp.509-522

Belge Tipi:

Makale

Yayın Tarihi:

3/2025

Belge Dili:

Türkçe

Document ID (Org No.):

2001334889

pISSN:

1308-9072