Types of Time Series Tasks

Overview of the different types of tasks when analyzing time series data

Updated: 03 September 2023

This document is a summary of this Overview of time series analysis Python packages

Analysis tasks

Forecasting

Forecast/predict future values based on some past data

Forecasting example

Classification

Classify time series data to groups/classes

Classificaion example

Clustering

Clustering series into similar groups based on statistical properties - different to classification since the groups are not predefined

Clustering example

Anomaly Detection

Anomaly/outlier/novelty detection is about finding abnormal:

  • Data points - outliers
  • Subsequences - discords

Anomaly detection example

Segmentation

Segmentation/summarization is about approximating time series data while retaining important features

Segmentation example

Pattern recognition

Pattern recognition/motif discovery is about finding subsequences that apper recurrently

Pattern recognition example

Indexing

Indexing is similar to pattern recognition, is query by content which is about finding sequences in a time series and can be used as the basis of other tasks like clustering and motifs discovery

Indexing example

Change point detection

Finding points in time where statistical properties like mean and variance change abruptly

Change point detection

Data preparation

Techniques used to support or imporve analysis

  • Dimensionality reduction
  • Missing value imputation
  • Decomposition
  • Preprocessing
  • Similarity measurement

Evaluation

Used for evaluating results of analysis tasks

  • Selection
    • Model selection
    • Hyperparameter search
    • Feature selection
  • Metrics and statistical tests
  • Visualization