Evaluating The Effects of Hyperparameter Tuning and Data Balancing on Machine Learning Algorithms Used for Heart Disease Prediction

Fuat Sungur; Halit Bakır

Makale detayları

Evaluating The Effects of Hyperparameter Tuning and Data Balancing on Machine Learning Algorithms Used for Heart Disease Prediction

Yazarlar:

Fuat Sungur | Halit Bakır

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Bilişim Teknolojileri Dergisi Vol.17 Issue 1, pp.45-58

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001310695

pISSN:

1307-9697

Anahtar kelimeler:

Özet:

 Neglecting the symptoms of heart disease can result in serious conditions and even death. Machine learning techniques can be used to make predictions about whether a person has heart disease based on these symptoms. In this study, heart disease prediction was performed using Logistic Regression, Decision Trees, Random Forest, K Nearest Neighbors, Naive Bayes, Gradient Boosting, XGBoost, and Bagging machine learning algorithms. Four separate datasets were created using data balancing methods such as SMOTE, SMOTETomek, Oversample Minority Class, and Undersample Majority Class. Hyperparameter optimization was conducted for all selected machine learning algorithms using Random Search and Bayesian Optimization techniques, and the results were compared. By comparing the impact of data balancing and hyperparameter optimization on the performance of machine learning techniques used in predicting heart disease, this study contributes to the literature with an original approach. The study utilized a dataset from a survey of 319,795 individuals in the United States, which included 20 relevant features. The Random Forest algorithm achieved a prediction accuracy of 94 in the model created using the SMOTETomek data balancing technique and Bayesian hyperparameter optimization. Additionally, the Random Forest algorithm, with the Oversample Minority Class data balancing technique and Bayesian hyperparameter optimization, achieved a classification accuracy of 97.

 Neglecting the symptoms of heart disease can result in serious conditions and even death. Machine learning techniques can be used to make predictions about whether a person has heart disease based on these symptoms. In this study, heart disease prediction was performed using Logistic Regression, Decision Trees, Random Forest, K Nearest Neighbors, Naive Bayes, Gradient Boosting, XGBoost, and Bagging machine learning algorithms. Four separate datasets were created using data balancing methods such as SMOTE, SMOTETomek, Oversample Minority Class, and Undersample Majority Class. Hyperparameter optimization was conducted for all selected machine learning algorithms using Random Search and Bayesian Optimization techniques, and the results were compared. By comparing the impact of data balancing and hyperparameter optimization on the performance of machine learning techniques used in predicting heart disease, this study contributes to the literature with an original approach. The study utilized a dataset from a survey of 319,795 individuals in the United States, which included 20 relevant features. The Random Forest algorithm achieved a prediction accuracy of 94 in the model created using the SMOTETomek data balancing technique and Bayesian hyperparameter optimization. Additionally, the Random Forest algorithm, with the Oversample Minority Class data balancing technique and Bayesian hyperparameter optimization, achieved a classification accuracy of 97.

 Neglecting the symptoms of heart disease can result in serious conditions and even death. Machine learning techniques can be used to make predictions about whether a person has heart disease based on these symptoms. In this study, heart disease prediction was performed using Logistic Regression, Decision Trees, Random Forest, K Nearest Neighbors, Naive Bayes, Gradient Boosting, XGBoost, and Bagging machine learning algorithms. Four separate datasets were created using data balancing methods such as SMOTE, SMOTETomek, Oversample Minority Class, and Undersample Majority Class. Hyperparameter optimization was conducted for all selected machine learning algorithms using Random Search and Bayesian Optimization techniques, and the results were compared. By comparing the impact of data balancing and hyperparameter optimization on the performance of machine learning techniques used in predicting heart disease, this study contributes to the literature with an original approach. The study utilized a dataset from a survey of 319,795 individuals in the United States, which included 20 relevant features. The Random Forest algorithm achieved a prediction accuracy of 94 in the model created using the SMOTETomek data balancing technique and Bayesian hyperparameter optimization. Additionally, the Random Forest algorithm, with the Oversample Minority Class data balancing technique and Bayesian hyperparameter optimization, achieved a classification accuracy of 97.

 Kalp hastalığı belirtilerinin ihmal edilmesi ciddi rahatsızlıklarla hatta ölümle sonuçlanabilir. Makine öğrenme teknikleri ile ön tanı için bu belirtiler kullanılarak kişide kalp hastalığı olup olmadığına dair tahmin yapılabilmektedir. Bu çalışmada Logistic Regression, Decision Trees, Random Forest, K Nearest Neighbors, Naive Bayes, Gradient Boosting, XGBoost ve Bagging algoritmaları ile kalp hastalığı tahmini yapılmıştır. SMOTE, SMOTETomek, Oversample Minority Class, Undersample Majority Class veri dengeleme yöntemleri ile dört ayrı veri seti oluşturulmuştur. Seçilen tüm makine öğrenme algoritmalarına Random Search ve Bayesian Optimizasyon teknikleriyle hiper parametre optimizasyonu yapılarak sonuçlar karşılaştırılmıştır. Veri dengeleme ve hiper parametre optimizasyonunun kalp hastalığının tahmininde kullanılan makine öğrenme teknikleri performansına etkisi karşılaştırılarak literatüre özgün bir çalışma kazandırılmıştır. Çalışmada Amerika Birleşik Devletlerinde 319.795 kişi ile yapılan 20 öz nitelikli bir anket olan veri seti kullanılmıştır. Random Forest algoritması SMOTETomek veri dengeleme tekniği kullanılarak ve Bayesian hiper parametre optimizasyonu yapılarak oluşturulan modelde 94 tahmin başarısı elde edilmiştir. Ayrıca, Random Forest algoritması ile Oversample Minority Class veri dengeleme tekniği kullanılarak ve Bayesian hiper parametre optimizasyonu yapılarak 97 sınıflandırma doğruluğu elde edilmiştir.

Makale detayları

Evaluating The Effects of Hyperparameter Tuning and Data Balancing on Machine Learning Algorithms Used for Heart Disease Prediction

Yazarlar:

Fuat Sungur | Halit Bakır

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Bilişim Teknolojileri Dergisi Vol.17 Issue 1, pp.45-58

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001310695

pISSN:

1307-9697

heart Diseases

random forest

machine learning

smote

smotetomek

bayesian optimizationKalp Hastalıkları

Random Forest

Makine Öğrenmesi

Smote

Smotetomek

Sayı

Makale detayları

Yazarlar: Fuat Sungur | Halit Bakır

Yayıncı: Gazi University

Ana Başlıklar: Makine Öğrenme (Diğer)

Dergi: Bilişim Teknolojileri Dergisi Vol.17 Issue 1, pp.45-58

Belge Tipi: Makale

Yayın Tarihi: 1/2024

Belge Dili: Türkçe

Document ID (Org No.): 2001310695

pISSN: 1307-9697

Sayı

Yazarlar:

Fuat Sungur | Halit Bakır

Yayıncı:

Gazi University

Ana Başlıklar:

Makine Öğrenme (Diğer)

Dergi:

Bilişim Teknolojileri Dergisi Vol.17 Issue 1, pp.45-58

Belge Tipi:

Makale

Yayın Tarihi:

1/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001310695

pISSN:

1307-9697