Machine Learning
Intermediate
Explainable AI & Interpretable Models
Uncover model predictions using SHAP values, LIME feature attribution, and partial dependence plots.
Course Overview
Deconstruct opaque black-box models into auditable, interpretable feature attributions for regulated industries like finance, healthcare, and risk assessment.
What You Will Learn
Compute global and local feature attributions using SHAP (Shapley Additive exPlanations)
Apply LIME (Local Interpretable Model-agnostic Explanations) to tabular and text data
Generate Partial Dependence Plots (PDP) and Accumulated Local Effects (ALE)
Audit black-box models for hidden bias, demographic disparity, and spurious correlations
Export human-readable model explanation reports for regulatory compliance
Tools & Technologies Used
SHAP
LIME
Scikit-Learn
XGBoost
Python 3.11
Matplotlib
Structured Curriculum
2 Modules · 4 Lessons · 150 Minutes Total
Module 1
Module 1: Game Theory & SHAP Feature Attributions
2 lessons
Module 1: Game Theory & SHAP Feature Attributions
Understand Shapley values and calculate exact local feature importance.
-
Shapley Value Foundations & TreeSHAP
Deconstruct cooperative game theory concepts applied to model prediction contributions.
-
Generating SHAP Summary & Force Plots
Visualize global feature interactions and individual decision breakdowns.
Module 2
Module 2: LIME & Algorithmic Fairness Audits
2 lessons
Module 2: LIME & Algorithmic Fairness Audits
Explain individual predictions locally and audit model equity.
-
LIME Explanations for Tabular & Text Classifiers
Fit surrogate linear models around complex predictions to generate intuitive explanations.
-
Auditing Bias & Disparate Impact
Measure demographic parity and equalized odds metrics across sensitive attributes.
Practical Project & Capstone Outcome
🚀 Capstone Project
Auditable Loan Application Risk Explainer
Train an XGBoost credit risk model and build an interactive explanation portal that provides applicants with clear, legally compliant reasons for decision outcomes.
Prerequisites
- Python intermediate proficiency
- Familiarity with Scikit-Learn or XGBoost
Intended Audience
- Data Scientists needing to explain complex model decisions to stakeholders
- Compliance & Risk Officers auditing algorithmic fairness in enterprise systems
Instructor Information
D
Deepak Sen
Course Author & Industry Expert
Deepak Sen is a Senior AI Audit Consultant who specializes in model explainability and regulatory compliance for financial institutions.
Frequently Asked Questions
Can SHAP be used with neural networks as well as tree-based models?
Yes! The course covers TreeSHAP for tree models and KernelSHAP/DeepSHAP for neural networks.