SHAP and Lime Deep Dive

This session explores SHAP and LIME as key methods for explainable AI in finance, linking their theoretical foundations to practical model interpretation.

0 Rating
0 Reviews
2 Students Enrolled

Suggested citation

Misheva, Branka Hadji (2025). SHAP and Lime Deep Dive. Q2 ecosystem. Available at https://quantinar.com/course/938/shap-and-lime-deep-dive or http://dx.doi.org/10.24818/IDA-CL/2025.143.

@misc{misheva2025, author = {Misheva, Branka Hadji}, title = {SHAP and Lime Deep Dive}, year = {2025}, howpublished = {Q2 ecosystem}, doi = {http://dx.doi.org/10.24818/IDA-CL/2025.143.}, url = {https://quantinar.com/course/938/shap-and-lime-deep-dive} }
Wishlist

Free

Enroll Now

Tags:

Courselet Content

1 components

	SHAP and Lime Deep Dive (pdf)	1.77 M

Requirements

1) Foundations in Machine Learning – understanding of supervised learning, model training, overfitting, feature importance. 2) Mathematical Background – basic linear algebra, probability, and statistics (enough to follow Shapley values and regression concepts). 3) Programming Skills – ability to work with Python (libraries such as scikit-learn, XGBoost, SHAP, LIME).

General Overview

Description

These slides, prepared within the MSCA Digital Finance program, provide a deep dive into explainable AI (XAI) methods in finance, focusing on SHAP (SHapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations).

The material begins with the theoretical foundations of Shapley values from cooperative game theory, explaining how fairness principles (efficiency, additivity, null player, symmetry) translate into quantifying feature importance in machine learning models. It then introduces practical applications of Shapley values through SHAP implementations such as KernelSHAP, LinearSHAP, DeepSHAP, and TreeSHAP.

The slides also cover LIME’s step-by-step algorithm, showing how it perturbs data, builds local surrogate models, and explains predictions. Comparisons between LIME and SHAP highlight differences in theoretical grounding, accuracy, and consistency.

Finally, the lecture touches on DeepLIFT and its role in explaining neural networks, before unifying explanation methods under the framework proposed by Lundberg and Lee (2017). Throughout, the focus is on making AI models in finance transparent, trustworthy, and interpretable.