|
Analysis of Factors Affecting Covid-19 Disease Using Patient Clinical Data by Data Mining Methods
|
Mansour Esmaeilpour   , Arash Bahiraei |
| Islamic Azad University - Hamadan Branch |
|
|
Abstract: (894 Views) |
Introduction & Purpose: The widespread infection of Covid-19 virus and its rapid spread in recent years has become one of the problems of medical staff in the world. When these patients enter medical centers and go through the initial stages of virus diagnosis, their optimal and regular separation from other patients can be of great help to the medical staff. On the other hand, faster diagnosis of a patient suspected of having Covid-19 will be effective in preventing its further spread. In this study, we intend to use the data of patient's medical images and data mining methods to determine the effective characteristics in diagnosing and differentiating patients with Covid-19 from other patients in medical centers. Also, decision-making rules have been extracted to optimize the decision-making process of the medical staff, especially in "better distinguishing patients from other patients."
Design/methodology/approach: In this research, we have used data mining with "ROUGH SET" method, "Artificial Neural Network" and "Decision tree" to extract decision rules. For this purpose, first the data was cleared and irrelevant features were removed, then using Excel, Rosetta, and Weka software, they attempted to dissect and divide the data into tests and training, and after reducing the features, extracting the rules of decision-making and analysis. That, action was taken.
Findings: In this study, a total of 950 radiographic images of patients are included, of which 311 lines or 32.7% are related to women and 559 lines or 58.8% are related to men. According to Johnson, Genetic and Decision tree, the characteristics of "age", "number of days of imaging after the onset of symptoms or hospitalization", "RT_PCR test status" and "type of medical image", and based on Holtz function plus the feature "oxygen demand", have a greater impact on diagnosis. Using the "ROUGH SET" method by Johnson, Genetics and Holtz methods, the accuracy of each model rule was 83%, which was extracted by Johnson (465), by genetic function (3316) and by Holtz (62). Using the "decision tree" method and its J48 algorithm, the model has an accuracy of 82% and a number of rules equal to 9. Also, in the "artificial neural network" method with the Perceptron multilayer algorithm, the model accuracy is approximately 97% higher than other methods. . On average, medical images were taken 9.08 days after the onset of symptoms and after hospitalization of patients. Finally, the five compound laws resulting from the implementation of these methods were explained. |
Article number: 1 |
| Keywords: Corona, Covid-19, Patient Location, Decision Tree, Rough, Neural Network, Montreal. |
|
|
Full-Text [PDF 1745 kb]
(380 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2022/01/23 | Accepted: 2022/12/15 | Published: 2023/01/1
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
