Time Series Analytics

Unit-03
Decomposition of Time Series

Semester-04
Master of Computer
Semester-02Application 1
 UNIT

Decomposition of Time Series

Names of Sub-Unit

Transformation and Adjustment for Real Data, Decomposing Data - Why Decompose?
Additive and Multiplicative Seasonality, Classical Decomposition. STL Decomposition,
Decomposition Model, Examples

Overview

This unit covers Transformation and Adjustment for Real Data, Decomposing Data - Why
Decompose? Additive and Multiplicative Seasonality, Classical Decomposition. STL
Decomposition, Decomposition Model, Examples

Learning Objectives

In this unit, you will learn to:

 Understand the concept of time series decomposition.

2
  Learn how to decompose a time series into trend, seasonality, and noise.
 Apply additive and multiplicative seasonality concepts to decompose time series
data.
 Use Python to implement classical time series decomposition techniques.
 Explore STL decomposition and its application in decomposing time series data.

Learning Outcomes

At the end of the unit you would

 Ability to transform and adjust real data for time series analysis.
 Understanding the reasons for decomposing time series data.
 Proficiency in identifying and applying additive and multiplicative seasonality
models.
 Competence in implementing classical decomposition techniques in Python.
 Capability to utilize STL decomposition for decomposing time series data, illustrated
through examples.

Pre-Unit Preparatory Material

 https://machinelearningmastery.com/decompose-time-series-data-trend-
seasonality/
 https://saturncloud.io/glossary/time-series-decomposition/

3
Table of Topics

3.1 Introduction to Time Series Decomposition

3.2 Transformation and Adjustment for Real Data
3.3 Importance of Decomposing Data
3.4 Additive and Multiplicative Seasonality Models
3.5 Classical Decomposition Techniques
3.6 Introduction to STL Decomposition
3.7 Implementation of Decomposition Models in Python
3.8 Examples of Time Series Decomposition
3.9 Summary

4
3.1 Introduction to Time Series Decomposition

Time series decomposition is a fundamental technique in time series analysis used to

understand the underlying patterns within a time series. It involves breaking down a time
series into its constituent components, which typically include trend, seasonality, and noise.
Trend represents the long-term direction of the series, indicating whether it is increasing,
decreasing, or stable over time. Seasonality captures the periodic fluctuations or patterns
that occur at regular intervals, such as daily, weekly, or yearly. Noise, also known as residual
or irregular component, represents the random fluctuations in the data that cannot be
attributed to the trend or seasonality.

The process of time series decomposition is valuable because it helps in separating the
different components of a time series, making it easier to analyze and interpret the data. By
decomposing a time series, analysts can better understand the underlying patterns and
trends, identify any seasonality effects, and assess the presence of any irregularities or
outliers in the data. This decomposition process is essential for various applications,
including forecasting, anomaly detection, and trend analysis, across a wide range of fields
such as finance, economics, and environmental science.

There are several methods for decomposing a time series, with two common approaches
being additive and multiplicative decomposition. In additive decomposition, the original
time series is decomposed into the sum of its components (trend, seasonality, and noise),
where each component is added together to reconstruct the original series. In contrast,
multiplicative decomposition involves decomposing the time series into the product of its
components, where each component is multiplied to reconstruct the original series. Both
approaches have their advantages and are used based on the nature of the data and the
specific requirements of the analysis.

3.2 Transformation and Adjustment for Real Data

Transformation and adjustment for real data in the context of decomposing a time series
involves preparing the data for analysis by addressing common issues such as non-
stationarity, seasonality, and outliers. This process is crucial as it ensures that the time series
decomposition is accurate and meaningful.

5
 Stationarity: Before decomposing a time series, it is essential to check for stationarity.
Stationarity implies that the statistical properties of the time series, such as mean and
variance, do not change over time. If the time series is non-stationary, it can be
transformed to achieve stationarity. Common transformations include taking the
logarithm, square root, or differencing the data.

 Seasonal Adjustment: Seasonality in a time series refers to regular, periodic

fluctuations that occur at fixed intervals. To adjust for seasonality, seasonal
decomposition techniques can be applied. These techniques estimate and remove the
seasonal component from the time series, allowing for a clearer view of the underlying
trend and noise.

 Outlier Detection and Treatment: Outliers are data points that significantly deviate
from the rest of the data. They can distort the analysis and affect the accuracy of the
decomposition. Outliers can be identified using statistical methods such as the Z-score
or by visual inspection. Once identified, outliers can be adjusted or removed from the
dataset to ensure they do not influence the decomposition process.

 Smoothing: Smoothing techniques can be applied to reduce the impact of noise in the
time series, making it easier to identify the underlying trend and seasonality. Common
smoothing techniques include moving averages and exponential smoothing, which
calculate a smoothed value for each data point based on neighboring data points.

 Data Scaling: In some cases, it may be necessary to scale the data to ensure that all
variables are on the same scale. This can be important when different components of
the time series have different magnitudes, making it difficult to compare them directly.
Scaling techniques such as min-max scaling or standardization can be used to scale the
data to a common range or standard deviation.

Transformation and adjustment for real data play a crucial role in preparing time series data
for decomposition analysis. By addressing issues such as non-stationarity, seasonality,
outliers, and noise, these techniques help ensure that the decomposition process is accurate
and produces meaningful results.

3.3 Importance of Decomposing Data

In time series analysis, transforming and adjusting real data is crucial to ensure the data

6
meets the assumptions of the decomposition models. Common transformations include
logarithmic, square root, and Box-Cox transformations. These transformations stabilize the
variance, make the data more symmetric, and help in handling outliers. Adjustment involves
dealing with irregularities in the data, such as missing values or outliers, which can affect the
accuracy of the decomposition. Techniques like interpolation, smoothing, or outlier
detection and removal are used to adjust the data for more reliable decomposition results.

In Python, the pandas library provides powerful tools for data transformation and
adjustment. For example, the log function can be applied to a time series to perform a
logarithmic transformation. Additionally, fillna can be used to handle missing values, and
rolling can be used for smoothing the data. Advanced techniques like robust statistics or
machine learning models can also be applied for more complex data transformations and
adjustments, depending on the nature of the data and the requirements of the analysis.

Importance of Decomposing Data

Decomposing time series data into its various components is essential for several reasons.
Firstly, it helps in understanding the underlying patterns and trends in the data, which is
crucial for making informed decisions in various fields such as finance, economics, and
environmental science. By decomposing the data, analysts can identify the long-term trends,
seasonal variations, and irregularities in the data, enabling them to better interpret the data
and extract meaningful insights.

Secondly, decomposing data is essential for forecasting future values of the time series. By
understanding the trend and seasonality components of the data, analysts can develop more
accurate forecasting models. For example, in a sales forecasting scenario, decomposing the
sales data can help in identifying the seasonal patterns, which can then be used to forecast
future sales more accurately, taking into account the seasonal variations.

Overall, decomposing data is a fundamental step in time series analysis that helps in
uncovering the underlying patterns and trends in the data, making it easier to interpret and
analyze the data, and enabling more accurate forecasting and decision-making.

7
3.4 Additive and Multiplicative Seasonality Models

Additive and multiplicative seasonality models are two common approaches used in time
series analysis to decompose seasonal components from the overall trend and noise in the
data.

Additive Seasonality Model:

In the additive seasonality model, the seasonal component is added to the trend and noise
components. Mathematically, the additive model can be represented as:

The additive model assumes that the seasonal fluctuations are consistent in magnitude over
time and are independent of the level of the time series.

Multiplicative Seasonality Model:

In the multiplicative seasonality model, the seasonal component is multiplied by the trend
and noise components. Mathematically, the multiplicative model can be represented as:

Where the terms are the same as in the additive model.

The multiplicative model assumes that the seasonal fluctuations are proportional to the level
of the time series. This means that as the level of the time series increases or decreases, the
seasonal fluctuations also increase or decrease proportionally.

8
Advantages and Considerations:

Additive Model:
 Easier to interpret: The additive model directly shows the seasonal component added
to the trend.
 Suitable for linear trends: Works well when the trend is relatively stable over time.

Multiplicative Model:
 Handles varying seasonal amplitudes: The multiplicative model is more flexible in
capturing seasonal patterns that change in magnitude over time.
 Suitable for exponential trends: Works well when the trend is growing or decaying
exponentially.

Implementation in Python:

Both additive and multiplicative seasonality models can be implemented using Python's
‘statsmodels’ library, specifically the ‘seasonal_decompose’ function. This function can
decompose a time series into its trend, seasonal, and residual components using either
additive or multiplicative models, depending on the specified parameter.
For example, to decompose a time series ts using an additive model:

9
Similarly, for a multiplicative model:

These models can provide valuable insights into the seasonal patterns present in the data,
helping in better understanding and forecasting of the time series.

3.5 Classical Decomposition Techniques

Classical decomposition techniques are traditional methods used in time series analysis to
separate a time series into its individual components, including trend, seasonality, and noise.
These techniques are foundational in understanding the underlying patterns within the data
and are often used as a starting point for more advanced analysis.

1. Moving Averages: Moving averages are a simple yet effective classical decomposition
technique. They involve calculating the average of a specific number of consecutive data
points, known as the window size. This moving average smooths out fluctuations in the data,
making it easier to identify the underlying trend. Common types of moving averages include
the simple moving average (SMA) and the exponential moving average (EMA).

2. Centered Moving Average: The centered moving average is a variation of the moving
average that places more emphasis on the central data points in the window. This helps in
reducing the lag effect of the moving average and provides a more accurate estimate of the
trend.

3. Exponential Smoothing: Exponential smoothing is another classical technique that

assigns exponentially decreasing weights to past observations. This means that more recent
observations are given greater weight than older observations. Exponential smoothing is
particularly useful for capturing short-term fluctuations in the data while still maintaining a
smooth trend estimate.

4. Holt-Winters Method: The Holt-Winters method, also known as triple exponential

smoothing, is an extension of exponential smoothing that incorporates seasonality into the

10
model. This method includes separate smoothing factors for the level, trend, and seasonal
components of the time series, allowing for more accurate decomposition of the data.

5. Seasonal Adjustment: Seasonal adjustment is a technique used to remove the seasonal

component from a time series. This can be done using methods such as seasonal
differencing, where the data is differenced by the seasonal period, or seasonal
decomposition of time series (STL), which decomposes the time series into trend, seasonal,
and residual components.

Advantages and Considerations:

 Simplicity: Classical decomposition techniques are relatively easy to implement and

understand, making them accessible to analysts with varying levels of expertise.

 Interpretability: The results of classical decomposition techniques are often easy to

interpret, providing clear insights into the underlying patterns within the data.

 Limited Flexibility: Classical decomposition techniques may not be suitable for time
series data with complex patterns or irregularities, as they are designed to capture
simple trends and seasonal components.

3.6 Introduction to STL Decomposition

STL (Seasonal-Trend decomposition using Loess) is a powerful decomposition technique

that separates a time series into its components: trend, seasonal, and residual. Unlike
classical decomposition methods, STL does not assume that the seasonal component has a
constant amplitude or frequency, making it more flexible and robust to irregularities in the
data.

1. Loess Smoothing: STL decomposition relies on the use of Loess (Locally Weighted
Scatterplot Smoothing) for estimating the trend and seasonal components. Loess is a non-
parametric regression technique that fits a smooth curve through the data by giving more
weight to nearby points and less weight to distant points.

2. Seasonal Adjustment: One of the key features of STL decomposition is its ability to
handle complex seasonal patterns. Unlike classical decomposition methods, which assume

11
a constant seasonal component, STL allows the seasonal component to vary over time. This
flexibility makes STL particularly useful for time series data with non-linear or changing
seasonal patterns.

3. Robustness to Outliers: Another advantage of STL decomposition is its robustness to

outliers. Loess smoothing is less sensitive to outliers compared to other smoothing
techniques, making STL more reliable in the presence of extreme values in the data.

4. Trend and Seasonal Extraction: In STL decomposition, the trend component represents
the long-term direction of the data, while the seasonal component captures the repeating
patterns that occur within a year or other fixed time interval. The residual component
represents the remaining variation in the data after the trend and seasonal components have
been removed.

Advantages and Considerations:

 Flexibility: STL decomposition can handle complex seasonal patterns and is not limited
to a constant seasonal amplitude or frequency.
 Robustness: The use of Loess smoothing makes STL decomposition robust to outliers
and irregularities in the data.
 Interpretability: The trend, seasonal, and residual components extracted by STL are
easy to interpret, providing valuable insights into the underlying patterns within the
data.
 Computationally Intensive: STL decomposition can be computationally intensive,
especially for large datasets, due to the need for multiple passes through the data for
trend and seasonal estimation.

12
3.7 Implementation of Decomposition Models in Python

Python offers several libraries for implementing time series decomposition models, such as
‘statsmodels’, ‘seasona’l, and ‘prophet’. These libraries provide easy-to-use functions for
decomposing time series data into its components, including trend, seasonality, and noise.

1. Using ‘statsmodels’ for Classical Decomposition:

‘statsmodels’ provides the ‘seasonal_decompose’ function to decompose a time series using

classical decomposition techniques. This function can be used to extract the trend, seasonal,
and residual components of a time series:

2. Using ‘STL’ for STL Decomposition:

For STL decomposition, ‘statsmodels’ provides the STL class, which uses Loess smoothing to
decompose a time series. Here's how you can use it:

3. Using ‘prophet’ for Time Series Decomposition:

‘prophet’ is a forecasting library developed by Facebook that includes a built-in function for

13
time series decomposition. Here's how you can use ‘prophet’ for decomposition:

Advantages and Considerations:

 Ease of Use: Python libraries such as statsmodels, seasonal, and prophet provide
simple and intuitive interfaces for implementing decomposition models.

 Visualization: These libraries also offer functions for visualizing the decomposed
components, making it easier to interpret the results.

 Flexibility: You can easily customize the decomposition models by adjusting

parameters such as the type of model (additive or multiplicative), the seasonal period,
and the smoothing parameters.

 Computational Efficiency: Python's efficient numerical libraries make it possible to

decompose large time series datasets quickly and effectively.

Python provides a wide range of tools and libraries for implementing time series
decomposition models, making it easier for analysts and data scientists to extract valuable
insights from time series data. By using these libraries, you can decompose time series data
into its components and gain a better understanding of the underlying patterns and trends.

3.8 Examples of Time Series Decomposition

Time series decomposition is a valuable technique for understanding the underlying

patterns within a time series dataset. Here, we provide examples of how to decompose a
time series into its trend, seasonal, and residual components using Python's ‘statsmodels’
library.

14
1. Example using Additive Model: In this example, we'll decompose a sample time series
dataset into its components using the additive model.

2. Example using Multiplicative Model: Similarly, we can decompose the same time series
dataset using the multiplicative model.

15
In these examples, we decompose a simple synthetic time series dataset into its trend,
seasonal, and residual components using both additive and multiplicative models. The
‘seasonal_decompose’ function from ‘statsmodels’ is used for decomposition, and the
resulting components are plotted for visualization

Time series decomposition is a useful technique for analyzing and understanding the
underlying patterns within a time series dataset. By decomposing a time series into its
components, analysts can gain insights into the trend, seasonal variations, and residual
fluctuations in the data, which can be valuable for forecasting and decision-making. Python's
‘statsmodels’ library provides easy-to-use functions for implementing time series
decomposition, making it accessible to analysts and data scientists.

3.9 Summary

 Transformation and adjustment of real data are essential for preparing time series data
for decomposition.

16
 Decomposing data helps in understanding underlying patterns, aiding in forecasting
and anomaly detection.

 Additive and multiplicative seasonality models offer different approaches to decompose

seasonal components.

 Classical decomposition techniques, such as moving averages and exponential

smoothing, provide simple yet effective ways to decompose time series data.

 STL decomposition is a robust technique that allows for flexible handling of seasonal
patterns.

 Loess smoothing is used in STL decomposition to estimate trend and seasonal

components.

 STL decomposition is advantageous for its robustness to outliers and ability to handle
complex seasonal patterns.

 Implementation of decomposition models in Python can be done using libraries like

statsmodels, seasonal, and prophet.

 Python provides tools for visualizing decomposed components, aiding in interpretation

of results.

 Examples demonstrate the practical application of decomposition techniques,

showcasing their utility in understanding time series data.

17
 Self- Assessment questions

A. Descriptive Type Questions

1. How does transformation and adjustment of real data impact the accuracy of time series
decomposition?

2. Explain the significance of decomposing data in time series analysis.

3. Compare and contrast additive and multiplicative seasonality models in terms of their
assumptions and applications.

4. What are the key features of classical decomposition techniques, and how are they
different from STL decomposition?

5. Can you provide an example of how STL decomposition can be used to analyze a real-
world time series dataset?

Answers for Self- Assessment questions

A. Hints for Essay Type Questions

1. Consider the effect of stabilizing variance and handling outliers.

2. Think about how decomposition helps in identifying trends, seasonality, and noise.
3. Focus on differences in assumptions about seasonal fluctuations.
4. Highlight differences in handling seasonal variations and robustness to outliers.
5. Consider discussing the steps involved in STL decomposition and its application to real
data.

18
 Post Unit Reading Material

 https://medium.com/analytics-vidhya/decomposition-in-time-series-data-
b20764946d63
 https://analyticsindiamag.com/why-decompose-a-time-series-and-how/

Topics for Discussion Forums

 Best practices for transforming and adjusting real data before decomposition in Python.
 Theoretical and practical reasons for decomposing time series data and its benefits in
Python.
 Comparing and contrasting additive and multiplicative seasonality models in Python.
 Real-world applications and case studies showcasing the effectiveness of STL
decomposition in Python.

19
20