Predictive Analytics (PA) for Improving Services
Predictive Analytics (PA) for Improving Services
Predictive Analytics for Improving Services
Overview
Defining PA for improving services
What is predictive analytics (PA)?
How can PA be used to improve services?
What are the limitations of PA for improving services?
What are the ethical considerations for using PA to improve services?
Is predictive analytics a good fit for your context?
Scoping a PA project
A motivating example
How will findings be used?
What are we trying to predict?
Who are we making predictions for?
When are we making predictions?
Conducting a PA proof-of-concept
The importance of a proof-of-concept
Defining learners
Specifying a benchmark learner
Adding and comparing learners
Data for PA: PART 1
Generalizing to new data
Data workflow
Identifying data
Designating data for testing
Training and validating learners
Additional considerations
Data for PA: PART 2
Data preparation
Overview of missing data
Handling missing data - predictors
Missing data - outcomes
Exploratory data analysies
Extracting predictive value
Changes in data over time
Modeling approaches
Overview
Logistic regression
Lasso
Decision trees and Random Forest
Gradient boosting
Naive Bayes
Support vector machines
Neural networks
Guidance for using ML
Evaluating and comparing predictive performance
The confusion matrix and threshold-variant metrics
Threshold-invariant metrics
Prioritizing metrics that align with goals
Selecting a threshold
Evaluating and comparing bias
Sources of bias
Bias in estimating predicting probabilities
Bias in how predicted probabilities are used
Getting specific about empirically defining bias
Context matters
Addressing bias
Understanding and communicating results
Understanding PA results
How are predicted probabilities distributed within and across groups?
Who has low or high predicted probabilities?
What are the most important predictors in my model?
Deploying predictive analytics
A brief overview of considerations
Additional optional material
Predictive Analytics vs. Inferential Statistics
Data model for code templates
Back to top