A Hybrid Model Focusing on Data Pre-Processing in Diabetes Diagnosis

Zeidi F., Azar L., Arslan V., EROL Ç.

CYBERNETICS AND SYSTEMS, vol.54, no.7, pp.1199-1211, 2023 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 54 Issue: 7
  • Publication Date: 2023
  • Doi Number: 10.1080/01969722.2022.2080338
  • Journal Name: CYBERNETICS AND SYSTEMS
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Applied Science & Technology Source, BIOSIS, Compendex, Computer & Applied Sciences, INSPEC, zbMATH
  • Page Numbers: pp.1199-1211
  • Keywords: Classification algorithms, diabetes diagnosis, hybrid model, K-means algorithm, normalization, outliers detection
  • Istanbul University Affiliated: Yes

Abstract

Diabetes mellitus is a common and serious disease that has been studied by many researchers. Pima Indians Diabetes Dataset is one of the most famous datasets in this field. This study aims to increase the accuracy of machine learning algorithms in diagnosing the disease and to reveal the patterns that enable early diagnosis of the disease by focusing on the pre-processing stages. The proposed hybrid model includes "filling in missing values with KNN", "examining six different normalization methods for normalization" and "removing outliers with K-means" in the pre-processing stage. In the data classification stage, four algorithms C4.5, SVM, Naive Bayes and KNN were examined and the best hybrid model was found. The performance evaluation of these models is based on accuracy. The results were compared with previous studies and had higher accuracy of 98.3% and 99.1% for (KNN + n5 + K-means + SVM) and (KNN + n4/n3 + K-means + KNN), respectively. Finally, we offer the conclusive notes and some suggestions for further study.