This book is a compilation of the most common mistakes when building machine learning models. I have gathered this list from mistakes I typically find when grading assignments, supervising graduate students, reading blog posts, looking at the accompanying code of published papers, and of course, from my own experience making those mistakes.

This book includes examples in Python. Some examples of mistakes that you will find in this book include:

- Not understanding the data

- Including irrelevant variables

- Data injection

- Assuming all users behave the same

- Wasting unlabeled data

- and much more!

Table of Contents

Introduction

Terminology

1 Not understanding the data

2 Reporting train performance

3 Not setting a seed value

4 Including irrelevant features

5 Ignoring differences in scales

6 Using the test set for fine tunning

7 Only reporting accuracy

8 Not comparing against a baseline

9 Not accounting for variance

10 Injecting data into the test set

11 Not shuffling the training data

12 Not saving the results 

13 Not parallelizing 

14 Encoding categories as integers

15 Forget data changes over time

16 Ignoring inter-user variance

17 Wasting unlabeled data

Apendix Setup Your Environment

🏆 The book is the WINNER of the Eric Ziegel Award 2023 which recognizes the best book reviewed in the journal Technometrics!

This is the leanpub version of the book. The most recent version is available for free online at ttps://enriquegit.github.io/behavior-free/ and the print version can be ordered here! In terms of content, all versions are very similar. The online and print versions include minor amendments and an additional comic illustration in the last chapter.

This book aims to provide an introduction to machine learning and deep learning concepts and algorithms applied to a diverse set of behavior analysis problems.

The book covers topics and practical aspects within the entire data analysis pipeline—from data collection, visualization, preprocessing, and encoding to model training and evaluation. No prior knowledge in machine learning is assumed. The book covers How To:

• Build supervised machine learning models to predict indoor locations based on Wi-Fi signals, recognize physical activities from smartphone sensors and 3D skeleton data, detect hand gestures from accelerometer signals, and much more.
• Apply some of the most common techniques to explore, visualize, encode, and preprocess behavioral data.
• Use unsupervised learning algorithms to discover criminal behavioral patterns.
• Program your own ensemble learning methods and use multi-view stacking to fuse signals from heterogeneous data sources.
• Encode your data using different representations, such as feature vectors, time series, images, bags of words, graphs, and so on.
• Train deep learning models with Keras and TensorFlow, including neural networks to classify muscle activity from electromyography signals and convolutional neural networks to detect smiles in images.
• Evaluate the performance of your models in traditional and multi-user settings.
• Train anomaly detection models such as Isolation Forests and autoencoders to detect abnormal fish trajectories.
• And much more!