Development of A Machine Learning Based Application For Diagnosing Autoinflammatory Diseases

Abstract

Diseases caused by immune system dysfunction have numerous complications, significantly affecting patients' life quality. One specific group of such diseases is autoinflammatory diseases, which occur when the immune system perceives certain tissues and organs as threats and attacks them. They are characterized by episodes of fever and high inflammation in joints and various tissues. Early diagnosis plays a crucial role in facilitating treatment and managing these disorders. Several factors have to be considered to ensure an accurate diagnosis. Along with signs and symptoms, comprehensive medical information, and usually confirmation via genetic tests, are required to diagnose autoinflammatory diseases. These procedures require a huge amount of resource utilization and result in prolonged diagnosis processes. In this thesis, we have developed a machine learning-based web application capable of accurately diagnosing Familial Mediterranean Fever (FMF), an autoinflammatory disease, without the need for genetic tests and with a minimal number of features. The process began with data acquisition, followed by analysis and a thorough understanding of the data's specifications and features. To enhance comprehension, data visualization techniques were employed. Subsequently, the data was cleaned, eliminating any noise, and manipulated to facilitate machine learning's grasp of the available data. Also, we conducted an extensive algorithm selection process to identify the most suitable algorithm for addressing the problem at hand. Then, a feature selection procedure was performed to minimize the number of features used for diagnosis. By analyzing the performance of multiple models, we identified the best-performing one, which was employed to build the diagnostic web application. We envision that the application will be helpful to physicians who seek expert consultation for diagnosis, minimizing the reliance on a huge number of patients' symptoms and the need for genetic tests.