Table Of ContentModern Time Series Forecasting
with Python
Explore industry-ready time series forecasting using modern
machine learning and deep learning
Manu Joseph
BIRMINGHAM—MUMBAI
Modern Time Series Forecasting with Python
Copyright © 2022 Packt Publishing
All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, without the prior written permission of the publisher, except in the case
of brief quotations embedded in critical articles or reviews.
Every effort has been made in the preparation of this book to ensure the accuracy of the information
presented. However, the information contained in this book is sold without warranty, either express
or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable
for any damages caused or alleged to have been caused directly or indirectly by this book.
Packt Publishing has endeavored to provide trademark information about all of the companies and
products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot
guarantee the accuracy of this information.
Publishing Product Manager: Dhruv Kataria
Senior Editors: Roshan Ravikumar, Tazeen Shaikh
Content Development Editor: Shreya Moharir
Technical Editor: Devanshi Ayare
Copy Editor: Safis Editing
Project Coordinator: Farheen Fathima
Proofreader: Safis Editing
Indexer: Subalakshmi Govindhan
Production Designer: Alishon Mendonca
Marketing Coordinator: Shifa Ansari
First published: November 2022
Production reference: 1181122
Published by Packt Publishing Ltd.
Livery Place
35 Livery Street
Birmingham
B3 2PB, UK.
ISBN 978-1-80324-680-2
www.packt.com
For my son, Zane,
For his boundless curiosity,
For his endless questions,
And for his innocent love of learning.
(All great qualities for adults who read this book as well.)
Contributors
About the author
Manu Joseph is a self-made data scientist with more than a decade of experience working with many
Fortune 500 companies, enabling digital and AI transformations, specifically in machine learning-
based demand forecasting. He is considered an expert, thought leader, and strong voice in the world
of time series forecasting. Currently, Manu leads applied research at Thoucentric, where he advances
research by bringing cutting-edge AI technologies to the industry. He is also an active open source
contributor and has developed an open source library—PyTorch Tabular—which makes deep learning
for tabular data easy and accessible. Originally from Thiruvananthapuram, India, Manu currently
resides in Bengaluru, India, with his wife and son.
About the reviewers
Dr. Julien Siebert is currently working as a researcher at the Fraunhofer Institute for Experimental
Software Engineering (IESE), in Kaiserslautern, Germany. He studied engineering sciences and AI
and obtained a PhD in computer science on the topic of modeling and simulation of complex systems.
After several years of research both in computer science and theoretical physics, Dr. Julien Siebert
worked as a data scientist for an e-commerce fashion company. Since 2018, he has been working at
the intersection between software engineering and data science.
Gerzson David Boros is the owner and CEO of Data Science Europe and a senior data scientist who
has been involved in data science for more than 10 years. He has an MSc and is a candidate for an
MBA. In the last 5 years, he and his team have made business proposals for 100 different executives
and worked on more than 30 different projects on the topic of data science and artificial intelligence.
His motto is “Social responsibility is also achievable with the help of data.”
Table of Contents
Preface xvii
Part 1 – Getting Familiar with Time Series 1
1
Introducing Time Series 3
Technical requirements 3 Stationary and non-stationary time series 13
What is a time series? 4 What can we forecast? 16
Types of time series 4 Forecasting terminology 17
Main areas of application for time series analysis 4
Summary 18
Data-generating process (DGP) 5 Further reading 18
Generating synthetic time series 7
2
Acquiring and Processing Time Series Data 19
Technical requirements 19 Slicing and indexing 26
Understanding the time series dataset 20 Creating date sequences and managing date
offsets 27
Preparing a data model 22
Handling missing data 28
pandas datetime operations,
Converting the half-hourly block-level data
indexing, and slicing
(hhblock) into time series data 33
– a refresher 23
Compact, expanded, and wide forms of data 33
Converting the date columns into
Enforcing regular intervals in time series 34
pd.Timestamp/DatetimeIndex 24
Converting the London Smart Meters dataset
Using the .dt accessor and datetime properties 25
into a time series format 35
viii Table of Contents
Mapping additional information 36 Hourly average profile 43
Saving and loading files to disk 38 The hourly average for each weekday 45
Seasonal interpolation 46
Handling longer periods of missing
data 38 Summary 48
Imputing with the previous day 42
3
Analyzing and Visualizing Time Series Data 49
Technical requirements 49 Detrending 60
Components of a time series 50 Deseasonalizing 61
Implementations 63
The trend component 50
The seasonal component 51 Detecting and treating outliers 72
The cyclical component 51 Standard deviation 72
The irregular component 51 Interquartile range (IQR) 73
Visualizing time series data 52 Isolation Forest 73
Extreme studentized deviate (ESD) and
Line charts 52
seasonal ESD (S-ESD) 73
Seasonal plots 55
Treating outliers 74
Seasonal box plots 56
Calendar heatmaps 58 Summary 75
Autocorrelation plot 58 References 75
Decomposing a time series 60 Further reading 75
4
Setting a Strong Baseline Forecast 77
Technical requirements 78 Seasonal naive forecast 83
Setting up a test harness 78 Exponential smoothing (ETS) 84
ARIMA 87
Creating holdout (test) and validation datasets 78
Theta Forecast 89
Choosing an evaluation metric 79
Fast Fourier Transform forecast 91
Generating strong baseline forecasts 80
Evaluating the baseline forecasts 94
Naïve forecast 82
Assessing the forecastability of a time
Moving average forecast 83
series 96
Table of Contents ix
Coefficient of Variation (CoV) 96 Summary 103
Residual variability (RV) 97 References 103
Entropy-based measures 98
Further reading 104
Kaboudan metric 101
Part 2 – Machine Learning for Time Series 105
5
Time Series Forecasting as Regression 107
Understanding the basics of machine Temporal embedding 116
learning 107
Global forecasting models – a
Supervised machine learning tasks 110 paradigm shift 116
Overfitting and underfitting 110
Summary 118
Hyperparameters and validation sets 113
References 119
Time series forecasting as regression 114
Further reading 119
Time delay embedding 114
6
Feature Engineering for Time Series Forecasting 121
Technical requirements 121 Exponentially weighted moving averages
(EWMA) 131
Feature engineering 122
Avoiding data leakage 123 Temporal embedding 134
Setting a forecast horizon 124 Calendar features 134
Time elapsed 135
Time delay embedding 125
Fourier terms 136
Lags or backshift 125
Rolling window aggregations 127 Summary 138
Seasonal rolling window aggregations 129
