مشخصات پژوهش

عنوان	تشخیص ناهنجاری در داده های مراقبت های بهداشتی با استفاده از یادگیری بدون نظارت
نوع پژوهش	پایان نامه
کلیدواژه‌ها	تشخیص ناهنجاری، داده های سلامت، یادگیری بدون نظارت
چکیده	The consequences of healthcare fraud and abuse are severe. However, detecting fraud and abuse in healthcare is difficult and limits auditing capabilities. This paper provides multiple unsupervised learning algorithms to recognize trends and detect aberrant healthcare interactions which can be fraudulent or abusive. Presented below are the study's most important conclusions, based on the anomaly detection model and its constituent models. We need precise labels identifying whether a certain person is aberrant or typical in order to compare the efficacy of the two techniques. A subset of 90 instances is chosen for this endeavor after the first execution of an isolation forest. The approach uses a sample's anomaly score to choose which ones to analyse. High and low anomaly scores, as well as a score close to the judgement score used to decide whether or not an observation is anomalous, are used to pinpoint the exact number of samples required (30 for each group). Business experts go at each scenario and identify it as "abnormal" or "normal," but they don't know which one will be expected. On the back of this research, we use five distinct algorithms: Isolation Forest, SVM, KNN, XGBoost, and DBSCAN. An in-depth look at the actual outcomes in the metrics for each approach demonstrates that XGBoost surpasses the competitors across the board.
پژوهشگران	رضا ندیمی (استاد مشاور)، علی ولی نژاد (استاد راهنما)، احمد وعد عبد علی (دانشجو)