No. 437 of 500 No. 74 in Hands-On AI & Programming

Feature Engineering for Machine Learning

Principles and Techniques for Data Scientists

By Alice Zheng and Amanda Casari

Published
May 8, 2018
Print length
215 pages
Reading level
Intermediate
ISBN-13
978-1491953242
Publisher
O'Reilly Media
Written for
Data scientists and ML engineers working with tabular or text data who want stronger feature intuition.

Synopsis

Feature engineering, the step of converting raw data into the numeric inputs from which a model learns, often decides whether a machine learning project succeeds, yet most textbooks treat it in passing. Alice Zheng and Amanda Casari give it a full book, built around one data problem per chapter. They begin with numeric data, covering filtering, binning, scaling, and log and power transforms, then move to text, where bag-of-words, n-grams, phrase detection and tf-idf weighting turn documents into vectors. Further chapters encode categorical variables with one-hot encoding, feature hashing and bin counting; reduce dimensionality with principal component analysis; use k-means clustering to create nonlinear features; and trace image features from hand-designed descriptors such as SIFT and HOG to representations learned by deep neural networks.

Each technique comes with Python code using NumPy, pandas, scikit-learn and Matplotlib, and the authors pair the math with intuition about what each transformation does to the data. A closing chapter applies several methods to a real structured dataset to build an academic paper recommender. Readers gain a vocabulary and toolkit for diagnosing why a model underperforms and for choosing transformations deliberately. Published in 2018, the book predates large language models and embedding-based pipelines, so its text and image chapters now serve as foundations rather than current practice.

Scorecard

Editorial score

How it scored

Editorial score: 67.3 / 1005 weighted criteria

  • Influence30%

    19.5/30
  • Authority25%

    17.5/25
  • Practical value20%

    15.0/20
  • Currency15%

    8.3/15
  • Readability10%

    7.0/10
  • Editorial score

    67.3/100

How the rank was computed

Rank No. 437 of 500 comes from a composite of 48.9: observable signals 18.5 of 55 points plus editorial review 30.4 of 45 points.

All 11 inputs
InputPoints
Observable signals18.5/55
Reach (Amazon ratings volume)7.1/16.5
Consensus across best-of lists5.5/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.2/8.3
Amazon sales rank0.7/5.5
Recency0.0/2.8
Editorial review30.4/45
Author authority7.9/11.3
Influence8.8/13.5
Practical value6.8/9.0
Readability3.2/4.5
Currency3.7/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A compact, intuition-first guide to turning raw data into model-ready features, still useful for tabular and text work despite its pre-LLM vintage.

Strengths

  • One focused data problem per chapter
  • Explains the intuition behind each transformation
  • Runnable Python examples with scikit-learn and pandas

Keep in mind

Its text and image chapters predate transformer embeddings and large language models, so practitioners in those areas will need newer references.

Key takeaways

  1. Simple transforms such as log scaling, binning and normalization can change model performance substantially, especially for linear models that are sensitive to scale.
  2. Tf-idf and related weighting schemes reshape bag-of-words vectors so that informative words carry more signal than common ones.
  3. High-cardinality categorical variables can be handled compactly with feature hashing or bin counting instead of enormous one-hot encodings.
  4. Model stacking, such as using k-means cluster membership as input features, lets simple linear models capture nonlinear structure in data.

About the author

Alice Zheng and Amanda Casari

Alice Zheng is a machine learning leader who has been a senior manager on Amazon's ad platform, director of data science at GraphLab/Dato/Turi and a researcher at Microsoft Research; she holds a Berkeley PhD in electrical engineering and computer science. Co-author Amanda Casari is a researcher and engineer in Google's Open Source Programs Office and External Faculty at the Vermont Complex Systems Center.

Questions

Is Feature Engineering for Machine Learning still relevant in 2026?
For tabular data, yes. Gradient-boosted trees and linear models on structured business data still depend on careful scaling, encoding and aggregation, and the book explains those choices clearly. Its text and image chapters are dated, since transformer embeddings and pretrained vision models have replaced most hand-built features there, so treat those chapters as background on why modern methods work.
Who should read Feature Engineering for Machine Learning?
Working data scientists, analysts moving into machine learning, and students who know basic Python and some linear algebra. It fits readers who can train a scikit-learn model but want to understand why their inputs help or hurt performance. Readers without any machine learning background should start with a general introduction, and deep learning specialists will find little new material.
What does Feature Engineering for Machine Learning cover?
It covers numeric transforms such as binning, scaling and log transforms; text representations including bag-of-words, n-grams and tf-idf; categorical encodings such as one-hot, feature hashing and bin counting; principal component analysis; k-means-based model stacking; and image features from SIFT and HOG to deep learning. A final chapter combines techniques to build an academic paper recommender in Python.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.