Time Series Analysis in Data Science: Forecasting Trends and Anomalies

Time series refers to a range of observations taken at a specific time period. For instance, a set of logins recorded for a regular period of each user. Similarly, time series data refers to a set of values arranged in the form of stock prices, sensor data, application telemetry, and click stream data. Anomaly is an unusual activity which is different from normal behavior. There are different methods adopted to find anomalies and trends in time series data in data science. Irrespective of technological advances in anomaly detection and system monitoring, there is still the risk of false positives. There are several anomaly detection models - some are generic while some are domain-specific. However, there is still a lack of a sufficient algorithm that can work with various datasets. Hence, this study will discuss various types of anomalies and components of time series data. This study will also propose a framework based on machine learning methods for detecting anomalies and selecting appropriate models for time series data analysis. This study will also present a taxonomy to define several aspects of “anomaly detection in time-series data. Performing anomaly detection regarding the context or types of activities a system is exposed to is one of the major challenges in existing techniques for anomaly detection. This study will help researchers to understand the emerging approaches of “time series anomaly detection” and computational methods.