Logistic Regression – An Applied Approach Using Python
Logistic Regression – An Applied Approach Using Python
The book is a showcase of logistic regression theory and application of statistical machine learning with Python. Topics include logit, probit, and complimentary log-log models with a binary target as well as multinomial regression. A section about contingency tables is also provided. Scikit-Learn and statsmodels are the two Python packages used to illustrate how to tune parameters for better fit and accuracy.
Table of Contents
1. Foreword
3-4
2. Binary Logistic Regression
5-19
Introduction
5
OLS Regression vs. a Problem with a Binary Outcome?
6
Odds and odds ratios
9
The logit model
11
Estimating the Logit Model
12
Scikit-Learn and Statsmodels API
13
3. Machine Learning and Binary Logistic Regression
20-45
Introduction
20
Evaluating Estimator Performance
20
Test Accuracy
22
ROC Curve
26
Precision-recall Curve
28
Issues with Convergence
28
Issues with Multicollinearity
30
Parameter Tuning
33
Parameter Tuning with GridSearch (GridSearchCV)
36
Additional Goodness-of-Fit Statistics
39
4. Probit and Complimentary Log-Log Models for Binary Regression
46-48
Introduction to Alternatives to Logit Models
46
Probit Mean Response Function
46
The Logistic Mean Response Function
47
Complementary Log-Log Response Function
47
5. Multinomial Logistic Regression
49-57
Introduction
49
Multinomial Logit
49
6. Contingency Tables
58-79
Introduction
58
Notation in Contingency Tables
59
Independence and Association
62
Cumulative Odds Ratios
68
Symmetry and Homogeneity
70
Working with 2X2 Tables
73
Stratified 2X2 Tables
75
Fisher Exact Test for Small Samples
78
