Abstract

Time series anomaly detection has ever existed as a fundamental analysis approach. The early time series anomaly detection techniques are mainly statistical and machine learning. For the practical processes of the deep neural network being constantly prospected by experimenters, the result of the deep neural network in anomaly detection tasks has been remarkably more helpful than conventional methods. Conventional models use commanded machine learning algorithms. In the proposed applications, organizing and annotating such a vast number of datasets is challenging, time-consuming, or too costly, and it needs specialization learning from professionals in the field. Hence, anomaly detection has become a significant challenge for investigators and practitioners. Anomaly detection is directed as the process of detecting anomaly data instances. In this analysis, we proposed an unsupervised and scalable framework for anomaly detection in time series data. The proposed technique is established on a variational auto-encoder. A deep, productive model that incorporates variational belief with deep learning. Also, real-time analysis has been performed for the time-series data. We used LSTM networks to process, make predictions, and classify based on time series data.