Sosyal ağ modelleme kullanilarak tibbi veri ağinda bağlanti tahmini yöntemleri

In the last few years we are witnessing to an increasing interest in the application of social network analysis and methods to health care information and management systems. Link prediction is an important task treated by social network analysis. For many different areas, link prediction can be used to expect future behavior or to recognize likely relationships that are hard or expensive to understand directly. One of these areas is related to medical care research area. Medical care area needs to become more proactive than reactive in recognizing the onset of disease and risk. Currently, physicians use laboratory results to further determine the patient's stage of health. However, there is some disadvantages such as generally to focus on only a few medical parameters (symptoms) and to linked by a particular doctor's experience, memory, and time. Therefore, current medical care is not proactive and is not enough treating or eliminating a disease at the earliest signs. As a remedy to the above mentioned problems, in this thesis, we propose a predictor to determine the risk of individuals to develop disease, and to undertake the correct actions at the earliest signs of illness. To this purpose, we first construct a weighted medical data network which indicates the relationships between abnormal parameters of disease. Then, we propose supervised and unsupervised link prediction methods based on the evolution of the constructed medical data network in order to identify the relations between all the parameters which can cause any disease, gathering the results obtained at several laboratories. Finally, we test the proposed method on the medical data network constructed with laboratory results of patients more than 210,000. In the next section of the thesis, a disease network considering the relations between diseases is proposed. Then, we present two link prediction methods based on supervised and unsupervised strategies to identify the connections between diseases, building the evolving structure of medical data network with respect to patients' ages. Experiments on a real network demonstrate that the proposed approach can reveal new abnormal parameter and disease correlations accurately and perform well at capturing future disease risks.