MeridianSpark

← HomeAll topicsToday's topic

July 18, 2026

literature-v3-2026-07-18July 18, 2026

Literature v3 · Research topic

Which AI explainer gives more consistent answers: LIME or SHAP?

Train a model to predict income, then run LIME and SHAP multiple times to see which one gives the same top features every time.

Data DetectiveAI EthicistReproducibility Champion
Zero-cost dataExploring two-variable data
8 weeksIntermediate$0 public datasets · Supplies: laptop onlyPortfolio 7/10

Why this matters

Imagine you train a model to predict income, but the explanation changes every time you run it. Which explainability method should you trust? This project pits LIME against SHAP in a reproducibility showdown on a classic census dataset.

Project scores

Originality7/10
Feasibility8/10
Impact7/10
Essay value6/10
Rigor6/10

Difficulty

Intermediate

This project is designed for high school students with some prior experience in Python and basic machine learning concepts. Over 8 weeks, you will learn to use LIME and SHAP libraries, run repeated experiments on a public dataset, and analyze variability in feature importance rankings. The workload is moderate, with weekly coding and analysis tasks. Prerequisites include familiarity with pandas, s

3 of 5 difficulty

Strengths

  • Clear comparison of two popular explainability methods
  • Reproducibility focus addresses a real research gap
  • Uses a well-known benchmark dataset (UCI Adult)
  • Quantitative analysis of stability across runs

Skills built

  • Python programming
  • Machine learning model interpretation
  • Statistical analysis of variability
  • Data visualization
  • Experimental design
  • Critical thinking about AI fairness

Zero-cost data

Zero-cost data

Research gap

Imagine you train a model to predict income, but the explanation changes every time you run it. Which explainability method should you trust? This project pits LIME against SHAP in a reproducibility showdown on a classic census dataset.

Curriculum alignment

Exploring two-variable dataData and informationScientific foundations of psychology

Unlock the full research scaffold & literature chain

Sign in to open competition fit, the complete milestone roadmap, admissions narrative drafts, and the full research write-up for this topic.

Full literature chainWeeks 3–12 milestonesAdmissions narrativeLaTeX & validation rubricsResearch plan