CONSTRUCTION AND STUDY OF A NEURO-FUZZY MODEL FOR DIAGNOSIS OF DIABETES MELLITUS
Authors:
1.
KNITU-KAI
2.
Kazanskiy nacional'nyy issledovatel'skiy tehnicheskiy universitet imeni A.N. Tupoleva
3.
Kazanskiy nacional'nyy issledovatel'skiy tehnicheskiy universitet imeni A.N. Tupoleva
4.
Kazanskiy nacional'nyy issledovatel'skiy tehnicheskiy universitet imeni A.N. Tupoleva
Type:
Article
Pages:
from 188
to 193
Status:
In work
Language:
Russian
Keywords:
DIABETES MELLITUS, DIAGNOSTICS, FUZZY NEURAL NETWORK, NEURO-FUZZY MODEL, DATA SET, KNOWLEDGE BASE FORMATION
Abstract (English):
The article describes the solution to the problem of constructing and studying a neuro-fuzzy model used to diagnose diabetes mellitus. Currently, there is a steady trend of increasing prevalence of diabetes mellitus. It is a group of endocrine diseases associated with impaired glucose absorption in the body. Timely diagnosis of this disease plays a key role in effective treatment and prevention of complications. The development of effective methods for diagnosing diabetes mellitus is an important task that requires an integrated approach using modern information technologies. To solve this problem, it is advisable to use fuzzy knowledge bases based on fuzzy-production rules formed using fuzzy neural networks. This approach is based on the principles of fuzzy logic, which allows you to obtain interpretable solutions. In this work, a fuzzy neural network model was used as a tool for forming knowledge bases, which has proven its effectiveness in solving a large number of problems in various subject areas. To train the fuzzy neural network, initial data was required, for which the Pima Indians dataset from the open source Kaggle was selected. The set includes 768 records for 9 parameters, one of which is an output parameter indicating the presence or absence of a disease. As a result of preprocessing of the initial data, 4 input parameters were selected that significantly affect the output parameter. The fuzzy neural network was trained using the software package "Neuro-fuzzy system for generating fuzzy models for assessing the discrete state of objects". Training and test samples were formed from the initial data using the group random sampling with replacement method. Triangular membership functions were used for training. During training, the fuzzy neural network went through 9 full iterations, which lasted a total of 55 minutes 32 seconds. For each iteration, the algorithm processed 38 parameters of the membership functions. For the training sample, the classification accuracy was 86.72%, for the test sample - 73.33%. During training, additional studies were conducted to assess the influence of the genetic algorithm parameters on the adequacy of the neuro-fuzzy model. The research results made it possible to formulate recommendations for choosing the training parameters of the fuzzy neural network. Comparison of the obtained results with known results of other authors allowed to conclude that the neuro-fuzzy model is effective for solving the problem of diabetes diagnosis. In the future, the constructed model can be integrated into an intelligent system and used in the diagnostic activities of doctors.
The article describes the solution to the problem of constructing and studying a neuro-fuzzy model used to diagnose diabetes mellitus. Currently, there is a steady trend of increasing prevalence of diabetes mellitus. It is a group of endocrine diseases associated with impaired glucose absorption in the body. Timely diagnosis of this disease plays a key role in effective treatment and prevention of complications. The development of effective methods for diagnosing diabetes mellitus is an important task that requires an integrated approach using modern information technologies. To solve this problem, it is advisable to use fuzzy knowledge bases based on fuzzy-production rules formed using fuzzy neural networks. This approach is based on the principles of fuzzy logic, which allows you to obtain interpretable solutions. In this work, a fuzzy neural network model was used as a tool for forming knowledge bases, which has proven its effectiveness in solving a large number of problems in various subject areas. To train the fuzzy neural network, initial data was required, for which the Pima Indians dataset from the open source Kaggle was selected. The set includes 768 records for 9 parameters, one of which is an output parameter indicating the presence or absence of a disease. As a result of preprocessing of the initial data, 4 input parameters were selected that significantly affect the output parameter. The fuzzy neural network was trained using the software package "Neuro-fuzzy system for generating fuzzy models for assessing the discrete state of objects". Training and test samples were formed from the initial data using the group random sampling with replacement method. Triangular membership functions were used for training. During training, the fuzzy neural network went through 9 full iterations, which lasted a total of 55 minutes 32 seconds. For each iteration, the algorithm processed 38 parameters of the membership functions. For the training sample, the classification accuracy was 86.72%, for the test sample - 73.33%. During training, additional studies were conducted to assess the influence of the genetic algorithm parameters on the adequacy of the neuro-fuzzy model. The research results made it possible to formulate recommendations for choosing the training parameters of the fuzzy neural network. Comparison of the obtained results with known results of other authors allowed to conclude that the neuro-fuzzy model is effective for solving the problem of diabetes diagnosis. In the future, the constructed model can be integrated into an intelligent system and used in the diagnostic activities of doctors.
Keywords:
DIABETES MELLITUS, DIAGNOSTICS, FUZZY NEURAL NETWORK, NEURO-FUZZY MODEL, DATA SET, KNOWLEDGE BASE FORMATION
DIABETES MELLITUS, DIAGNOSTICS, FUZZY NEURAL NETWORK, NEURO-FUZZY MODEL, DATA SET, KNOWLEDGE BASE FORMATION
