A Hybrid Outlier Detection Method for Health Care Big Data

Technology advancements in health care informatics, digitalizing health records, and telemedicine has resulted in rapid growth of health care data. One challenge is how to effectively discover useful and important information out of such massive amount of data through techniques such as data mining. Outlier detection is a typical technique used in many fields to analyze big data. However, for the large scale and high dimensional heath care data, the conventional outlier detection methods are not efficient. This paper proposes a novel hybrid outlier detection method, namely, Pruning-based K-Nearest Neighbor (PB-KNN), which integrates the density-based, cluster based methods and KNN algorithm to conduct effective outlier detection. The proposed PB-KNN adopts the case classification quality character (CCQC) as the medical quality evaluation model and uses the attribute overlapping rate (AOR) algorithm for data classification and dimensionality reduction. To evaluate the performance of the pruning operations in PB-KNN, we conduct extensive experiments. The experiment results show that the PB-KNN method outperforms the k-nearest neighbor (KNN) and local outlier factor (LOF) in terms of the accuracy and efficiency.