Explainable AI: The Power of Interpreting ML Models

Explainable machine learning, or XAI, aims to interpret machine learning models to uncover the most influential predictors of an outcome, ultimately enhancing transparency in predictive analytics. The primary goal of this project is to employ explainable machine learning techniques to explain the decision-making process of three distinct models: Logistic Regression, Random Forest, and Neural Networks. These three models were chosen to show three distinct ways of explaining a model, namely intrinsic, feature importances, and permutation importances. By working with the UCI Census Income dataset, we aim to predict whether an individual earns more than $50k/year.

This project also shows the difference between global and local explanation methods. Whereas global explanations show the most important predictors on average for the entire group, local methods aim to explain why a model made a prediction for an individual. Local methods are valuable for cases such as a client being rejected a loan—the company could provide an explanation of why the client was rejected. The client then has specific knowledge on what to change to improve their chances of obtaining a loan. For the local explainable methods, we focus on applying SHAP or SHapley Additive exPlanations.