Introduction to Predictive Analytics in Python
Build a logistic regression model in Python, choose its variables, and explain it to business stakeholders with gains and lift graphs.
Logistic regression sits at the center of this course: you use it to predict a yes-or-no target, such as whether a donor will give, from a base table of candidate variables. A model only gets used when people can follow it, so the lessons pair building the model with picking meaningful variables, measuring how well its predictions hold up and showing business stakeholders why each variable matters.
What You Will Achieve
- Predict a binary target in Python and pick out the new examples most likely to donate
- Compare models built on different sets of variables and decide how many variables the final model keeps
- Check a model on data held out from training and set a sensible cut-off for its predictions
- Use graphs to show business stakeholders how the model beats random targeting and why its variables make sense
Program Overview
Building Logistic Regression Models
Sets up the base table and fits a model that predicts a binary target from continuous variables, then uses it to score new examples.
- Base table structure and predictive variables
- Interpreting coefficients and the intercept
- Making predictions for new donors
Forward Stepwise Variable Selection
Shows why a careful choice of variables makes a more useful model and how to add them one at a time while checking performance on separate data.
- Motivation for variable selection
- AUC and correlated variables
- Train and test partitioning
- Choosing the number of variables and the cut-off
Explaining Model Performance and Variables
Turns model results into graphs that business stakeholders can act on and explains how each variable in the model relates to the target.
- Cumulative gains and lift curves
- Business cases for targeting
- Predictor insight graphs
- Discretizing continuous variables
Certificate Awarded (Paid)
Job Outlook
Predictive models like these support targeting decisions, such as choosing which donors to contact, and the graphs help analysts explain those models and their value to business stakeholders.
Good for: Python users who want to build a first predictive model and learn to explain and defend it in front of business stakeholders.
Keep in mind: Every chapter works with one model type, logistic regression for a binary target, so the depth goes into evaluating and explaining that single model.
Bottom line: A logistic regression course that gives two of its four chapters to gains curves, lift and predictor insight graphs for convincing business users.
Frequently Asked Questions
What is forward stepwise variable selection?
It is a way of building a model by adding variables one at a time, each step picking the next best variable. The lessons use it to find the order of variables and then decide how many the final model should keep.
What are predictor insight graphs?
They show how a single variable relates to the target. Continuous variables are first discretized into groups, then the graph plots the average incidence of the target and the size of each group, which helps explain the model's logic to others.
How is model performance measured in the lessons?
Models are compared by AUC, calculated on separate train and test partitions. Later chapters add cumulative gains and lift curves, read against a random model and a perfect model, to show how much better the model targets.


There are no reviews yet.