You're reading from XGBoost for Regression Predictive Modeling and Time Series Analysis Learn how to build, evaluate, and deploy predictive models with expert guidance

Product type Paperback

Published in Dec 2024

Publisher Packt

ISBN-13 9781805123057

Length 308 pages

Edition 1st Edition

Concepts

Data Governance

Authors (2):

Joyce Weiner

Partha Pritam Deka

View More author details

Table of Contents (19) Chapters

Preface

1. Part 1:Introduction to Machine Learning and XGBoost with Case Studies

2. Chapter 1: An Overview of Machine Learning, Classification, and Regression FREE CHAPTER

3. Chapter 2: XGBoost Quick Start Guide with an Iris Data Case Study

4. Chapter 3: Demystifying the XGBoost Paper

5. Chapter 4: Adding on to the Quick Start – Switching out the Dataset with a Housing Data Case Study

6. Part 2: Practical Applications – Data, Features, and Hyperparameters

7. Chapter 5: Classification and Regression Trees, Ensembles, and Deep Learning Models – What’s Best for Your Data?

8. Chapter 6: Data Cleaning, Imbalanced Data, and Other Data Problems

9. Chapter 7: Feature Engineering

10. Chapter 8: Encoding Techniques for Categorical Features

11. Chapter 9: Using XGBoost for Time Series Forecasting

12. Chapter 10: Model Interpretability, Explainability, and Feature Importance with XGBoost

13. Part 3: Model Evaluation Metrics and Putting Your Model into Production

14. Chapter 11: Metrics for Model Evaluations and Comparisons

15. Chapter 12: Managing a Feature Engineering Pipeline in Training and Inference

16. Chapter 13: Deploying Your XGBoost Model

17. Index

Why subscribe?

18. Other Books You May Enjoy

Understanding the XGBoost algorithm

In this section, you will learn how the XGBoost algorithm tackles problems with current basic gradient-boosted tree algorithms. You will cover the improvements the authors highlight in the paper and how the improvements help correct problems. First, you will learn about how the authors addressed problems with data, then you will learn about the improvements in XGBoost that speed up training.

Addressing problems – sparse data, overfitting

To handle overfitting, a change the authors made from the standard gradient-boosted tree algorithm is to add a function (Ω, called omega in the paper) for the complexity of the model. This function smooths the weights to avoid overfitting. The omega function does this by penalizing complexity, meaning the algorithm prefers solutions that are simpler. This function also makes the algorithm easier to parallelize for faster computation.

Two additional techniques to handle overfitting are used:

...