Paper: Explaining ML predictions with SHAP by ab93 · Pull Request #1104 · scipy-conference/scipy_proceedings

ab93 · 2025-06-13T22:13:45Z

If you are creating this PR in order to submit a draft of your paper, please name your PR with Paper: <title>. An editor will then add a draft label; this will trigger GitHub Actions to run automated checks on your paper and build a preview. You may then work to resolve failed checks and ensure the preview build looks correct. If you have any questions, please tag the proceedings team in a comment on your PR with @scipy-conference/2025-proceedings.

See the project readme for more information.

Signed-off-by: Avik Basu <avik_basu@intuit.com>

github-actions · 2025-06-13T22:14:08Z

Curvenote Preview

Directory	Preview	Checks	Updated (UTC)
papers/avik_basu	🔍 Inspect	✅ 37 checks passed (5 optional)	Oct 14, 2025, 6:06 PM

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

Signed-off-by: Avik Basu <avik_basu@intuit.com>

ameyxd · 2025-07-02T19:25:02Z

@ameyxd will serve as editor for this paper.

ameyxd · 2025-08-06T19:31:25Z

We are swiftly assigning new reviewers to this paper.

ameyxd · 2025-08-20T22:07:25Z

Assigned new reviewers, yet to respond - Shubham Sharma, LindseyCster. In process to reassign reviewers.

ameyxd · 2025-08-20T22:51:08Z

I will be joining as reviewer @ab93.

ameyxd

Generally well-written, accessible content with some room for improvement. I suggested some changes here.

ameyxd · 2025-08-20T22:53:04Z

+The Bank Marketing dataset [@bank_marketing_222] contains information about direct marketing campaigns (phone calls)
+of a Portuguese banking institution. It contains customer demographic information, financial details and interation
+history. The goal is to predict whether a customer will subscribe to a term deposit [@Moro2014ADA]. The dataset is 
+available on the UCI Machine Learning Repository.


Recommend citing the link to the UCIMLR here.

ameyxd · 2025-08-20T22:54:20Z

+
+### SHAP Explanations
+
+We use the `shap` Python library [@lundberg2020local2global] to compute the SHAP values for the XGBoost model.


I recommend expanding on what this section is for here.

ameyxd · 2025-08-20T22:55:45Z

+In terms of insights, the beeswarm plot {numref}`fig:beeswarm-xgb` shows some interesting observations. For example,
+customers with more balance (`balance` feature) in their account are more likely to subscribe to a term deposit. 
+On the other hand, customers who have been contacted more frequently in the current compaign (`campaign` feature) 
+are less likely to subscribe to a term deposit. However, the plot is not able to convey a deeper insight into the


Provide a reasoning/suggestion why the categorical features are not showcased correctly through SHAP, and the need to look at dependency plots here.

ameyxd · 2025-08-20T22:57:03Z

+   additional feature engineering.
+
+3. While SHAP values quantify the impact of features on model predictions, it is important to understand that they 
+   do not directly indicate causality.


Expand on why causality is not indicated by SHAP values in summary here.

ab93 · 2025-08-21T00:25:19Z

Generally well-written, accessible content with some room for improvement. I suggested some changes here.

@ameyxd thank you for the review! Will revise accordingly!

ameyxd · 2025-08-24T19:24:22Z

@ab93 - checking in on this. Please complete the edits ASAP. Failure to do so by EOW might result in this work being rejected from the proceedings.

ab93 · 2025-08-24T19:42:24Z

@ameyxd will do so by EOD today.

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

ab93 · 2025-08-24T20:43:41Z

@ameyxd let me know if the changes look good or if there are any further changes you need. Will fix them ASAP today.

ameyxd · 2025-08-25T13:47:22Z

LGTM. Good job. Approving now. @ab93

Avik Basu added 22 commits May 28, 2025 22:26

chore: first commit

af7127e

Signed-off-by: Avik Basu <avik_basu@intuit.com>

docs: update myst

ee31017

Signed-off-by: Avik Basu <avik_basu@intuit.com>

abstract and more bibtex

fb22152

Signed-off-by: Avik Basu <avik_basu@intuit.com>

improve introduction

ce4b265

Signed-off-by: Avik Basu <avik_basu@intuit.com>

first version of introduction

d474515

Signed-off-by: Avik Basu <avik_basu@intuit.com>

starting core concepts section

2fcdd78

Signed-off-by: Avik Basu <avik_basu@intuit.com>

complete core concepts

1fd3153

Signed-off-by: Avik Basu <avik_basu@intuit.com>

start with SHAP for GBDTs

ce69ad4

Signed-off-by: Avik Basu <avik_basu@intuit.com>

Add first plots in GBDT

fdb9406

Signed-off-by: Avik Basu <avik_basu@intuit.com>

explainations for gbdt shap

100a066

Signed-off-by: Avik Basu <avik_basu@intuit.com>

start with CNN analysis

d70dfec

Signed-off-by: Avik Basu <avik_basu@intuit.com>

more code snippets for gbdt

426894d

Signed-off-by: Avik Basu <avik_basu@intuit.com>

start with model definition

c154824

Signed-off-by: Avik Basu <avik_basu@intuit.com>

add cnn model class

d1693e7

Signed-off-by: Avik Basu <avik_basu@intuit.com>

details about model training

7626690

Signed-off-by: Avik Basu <avik_basu@intuit.com>

add conclusion

d222b22

Signed-off-by: Avik Basu <avik_basu@intuit.com>

completed strengths and limitations

6b5cc60

Signed-off-by: Avik Basu <avik_basu@intuit.com>

add perf metrics for cnn

ec2aa53

Signed-off-by: Avik Basu <avik_basu@intuit.com>

add global and dep plots

b6c19f0

Signed-off-by: Avik Basu <avik_basu@intuit.com>

added all plots

2eb0f31

Signed-off-by: Avik Basu <avik_basu@intuit.com>

add content for global shap in cnn

d90c958

Signed-off-by: Avik Basu <avik_basu@intuit.com>

revert back unwanted changes

470fa8e

Signed-off-by: Avik Basu <avik_basu@intuit.com>

rowanc1 added paper This indicates that the PR in question is a paper draft This triggers Curvenote Preview actions labels Jun 13, 2025

ab93 added 4 commits June 15, 2025 20:29

fix: add CRediT roles

c6f01fb

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

fix: add doi

3921ec4

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

fix: ignore lightgbm doi

f778450

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

fix: lightgbm citation

34696c3

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

scipy-conference deleted a comment from github-actions Bot Jun 18, 2025

ameyxd added the assigned-reviewer label Jun 23, 2025

Avik Basu added 2 commits June 25, 2025 12:29

feat: add notebook for GBDT

0fd6743

Signed-off-by: Avik Basu <avik_basu@intuit.com>

fix: setminus error

991e1fa

Signed-off-by: Avik Basu <avik_basu@intuit.com>

This comment was marked as resolved.

Sign in to view

chore: reformat gbdt nb

928fb21

Signed-off-by: Avik Basu <avik_basu@intuit.com>

This comment was marked as resolved.

Sign in to view

Avik Basu added 3 commits June 26, 2025 22:47

feat: first version of cnn nb

7ae7d0f

Signed-off-by: Avik Basu <avik_basu@intuit.com>

fix: affiliations

03a5909

Signed-off-by: Avik Basu <avik_basu@intuit.com>

feat: improve reproducibility

626914f

Signed-off-by: Avik Basu <avik_basu@intuit.com>

ameyxd self-assigned this Jul 2, 2025

ameyxd added the assigned-editor label Jul 2, 2025

improve content for Dependency and Local plots

71c566a

ameyxd reviewed Aug 20, 2025

View reviewed changes

ab93 added 5 commits August 24, 2025 13:03

comments: add UCIMLR citation

2f8c9da

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

comments: address expanding shap explanation section

02ca7aa

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

comments: reason why categorical features need dependency plots

fac0fde

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

comments: causality reason

05ed180

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

fix: doi error

fc0b2f2

Signed-off-by: Avik Basu <ab93@users.noreply.github.com>

fwkoch added approved This triggers Curvenote Submission action and removed draft This triggers Curvenote Preview actions labels Oct 14, 2025

fwkoch merged commit 5174ea9 into scipy-conference:2025 Oct 14, 2025
16 checks passed


		### SHAP Explanations

		We use the `shap` Python library [@lundberg2020local2global] to compute the SHAP values for the XGBoost model.

Uh oh!

Conversation

ab93 commented Jun 13, 2025

Uh oh!

github-actions Bot commented Jun 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as resolved.

This comment was marked as resolved.

ameyxd commented Jul 2, 2025

Uh oh!

ameyxd commented Aug 6, 2025

Uh oh!

ameyxd commented Aug 20, 2025

Uh oh!

ameyxd commented Aug 20, 2025

Uh oh!

ameyxd left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ab93 commented Aug 21, 2025

Uh oh!

ameyxd commented Aug 24, 2025

Uh oh!

ab93 commented Aug 24, 2025

Uh oh!

ab93 commented Aug 24, 2025

Uh oh!

ameyxd commented Aug 25, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

github-actions Bot commented Jun 13, 2025 •

edited

Loading