Exploring asymmetric pruning evolution for detecting anomalies in health monitoring time series

The healthcare datasets are used in machine intelligence-based systems to predict the ailments. The robotic surgeries and robotic procedures also use the healthcare datasets for training and for trials also. The anomalies in the dataset lead to erroneous predictions as the dataset works as input to the machine learning algorithms. The healthcare decisions should be taken timely to save the lives of patients, and the decisions taken by machine intelligence are always dependent upon the healthcare data. It is imperative to detect anomalies in the healthcare monitoring data, so that the machines can give the right output while diagnosing the human health on the basis of biomarker values stored in the healthcare datasets. Hence, this research work is attempting to find a solution using deep learning as a sub-set of machine learning to detect anomalies from the health monitoring data. Anomaly detection algorithms have traditionally relied on specialized error and rule sets to identify anomalous classes. Although significant advances have been achieved with unsupervised techniques like deep learning and sample reconstruction, these approaches still need more structure to accurately describe inputs and outputs across time. In this article, we demonstrate how to build a predictive architecture for anomaly detection using an asymmetric pruning evolution neural network. By studying the behaviour of time series in isolation, an attempt is made to identify unique patterns that precede anomalies and provide clues about what may follow. By training the algorithm to prioritize historical information, we are able to detect anomalies in the health monitoring data. The state-of-the-art performance of our technique has been experimentally validated across a variety of health monitoring datasets.