TLDR: Time Series Smoothing for Anomaly Detection

Date: 2020-11-01 Source: https://arpitbhayani.me/blogs/ts-smoothing

Overview

Learn how to optimally smooth time series data using kurtosis to highlight anomalies. Prioritize user attention with this smoothing technique. Time series is a collection of numerical data points (often measurements), gathered in discrete time intervals, and indexed in order of the time.

Key Points

• Time series is a collection of numerical data points (often measurements), gathered in discrete time intervals, and indexed in order of the time.
• Aberrations and Anomalies: In any data distribution, the anomalies and aberrations form in the long tail which means they are some extreme values that are far away from the mean.
• Kurtosis: Kurtosis is the measure of “tailedness” of the probability distribution (data distribution) and it helps in describing the shape of the plot.
• Finding the Optimal Window Size: As established earlier, anomalies and aberrations are extreme values that largely deviate from the mean and hence occupy a position on either tail of the distribution.