Detection of anomalies in data of pediatric congenital heart disease ICU

Congenital heart disease (CHD) is one kind of cardiovascular disease among children. Despite remarkable success in the surgical and medical management, the complications and mortality risk in ICU is still very high. A prediction model taking into account the associated risk parameters in the ICU can play a vital role in the healthcare domain to establish an early diagnosis of a severe condition. However, to develop such a successful model using machine learning techniques, the high-quality data, i.e., data without “anomalies” or outliers, is essential. This study analyzes the detection of anomalies in postoperative ICU data to accurately forecast patient states using several algorithms. We first introduced three anomaly detection algorithms, then assessed their effectiveness and selected the most effective anomaly detection algorithm. The experimental findings demonstrate that the most recent Local Outlier Factor (LOF) algorithm performs effectively on this job and yields promising outcomes.