Detection of Anomalies in Time Series Data

Abstract

Time Series are the observations that are noted in regular manner and are correlated with time. They help to extract meaning and summary of dataset and are arranged in chronological order. Time series data help us to analyze from past behavior and predict future behavior as well. Some of Time series data which are available are data from satellite, ola cab rides, stock market, online transactions etc. Now-a-days determining anomalies in time series is a major area of interests among researchers. Data is growing rapidly in every area. For example every hour stock market price will get updated and handling such a huge data become difficult for humans. If the anomaly is not detected, people will invest in wrong business lose their money. In this study deep learning models like ARIMA, Isolation forest, and LSTM based autoencoders are used to detect the anomalies in the dataset. Here stock market dataset is used to check whether the closing price at the end of the day is correct value or not. The dataset is preprocessed and then send to any one of the above mentioned model and an analysis in the performance is done between the models.