Capstone focuses on meta-learners and demonstrates an end-to-end uplift algorithm validation framework
The project explores cutting-edge uplift modeling techniques. Uplift modeling is a set of data science techniques that leverage both causal inference and machine learning to predict the effect of a treatment. It can be used to understand if a treatment (such as a marketing advertisement) caused the desired outcome (such as a purchase). The end result is typically a scored list of customers to prioritize because they can be persuaded by the intervention.
The project contains both research and application. In result, the project is presented as two components:
- An in-depth review of uplift modeling and its applications
- A deep-dive on a special set of uplift methods available in Python called meta-learners
- Senior Capstone Paper (includes all references)
- Slides giving an overview of project
- Jupyter notebooks with all work included
- Basic_Opossum - creation of the basic dataset
- Basic_EDA_Preprocessing_FeatureSelection - EDA, Preprocessing, Feature Selection for basic dataset
- Basic_TwoModel_ModelSelection_PerformanceEvaluation - Independent Two-Model Classifer Approach for basic dataset
- Basic_Pylift - Pylift Tuned XGB Approach for basic dataset
- Basic_CausalML - Meta-learner exploration (no R-learner) for basic dataset
- Complex_Opossum - creation of the basic dataset
- Complex_EDA_Preprocessing_FeatureSelection - EDA, Preprocessing, Feature Selection for basic dataset
- Complex_TwoModel_ModelSelection_PerformanceEvaluation - Independent Two-Model Classifer Approach for basic dataset
- Complex_Pylift - Pylift Tuned XGB Approach for basic dataset
- Complex_CausalML - Meta-learner exploration (no R-learner) for basic dataset