Performance of neural networks and heuristic models for disease prediction from liver enzymes: Application to biochemistry device output Karaciğer enzimlerinden hastalık tahmini için yapay sinir ağları ve sezgisel yöntem modellerinin performansları: Biyokimya cihazı çıktılarına uygulanması


Journal of the Faculty of Engineering and Architecture of Gazi University, vol.39, no.4, pp.2263-2270, 2024 (SCI-Expanded) identifier

  • Publication Type: Article / Article
  • Volume: 39 Issue: 4
  • Publication Date: 2024
  • Doi Number: 10.17341/gazimmfd.1268957
  • Journal Name: Journal of the Faculty of Engineering and Architecture of Gazi University
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Art Source, Compendex, TR DİZİN (ULAKBİM)
  • Page Numbers: pp.2263-2270
  • Keywords: Artificial neural networks, Liver diseases, Logistic regression, Particle swarm optimization
  • Istanbul Gelisim University Affiliated: Yes


In the application of decision-making systems in the field of healthcare, with advancing technology, the outputs of direct analysis devices have become usable. As the dataset becomes richer, the accuracy of models also increases. The parameters of the dataset used in this study contain raw data closer to real conditions in terms of both quantity and quality compared to previous studies. When examining the models established to identify liver diseases, it is observed that besides the model performance, the performance of experts also affects due to the use of parameters containing expert opinions. The data set used in this study did not include subjective data other than class values, and only expert opinions were used in training the model. Thus, the model performance will be less dependent on the dataset compared to other studies. Real-life data has been worked on with different models to see which structures are better. Artificial neural networks and particle swarm optimization methods were trained to solve the classification problem and results were analyzed by testing with training and test data in the study.