casual-inference

[TNNLS-2025, arXiv-2023.2.10] Official repository of "A Survey on Causal Reinforcement Learning"

Articles/ Journals and Videos related to Economics:chart_with_upwards_trend: and Data Science :bar_chart:

Implementation of paper DESCN, which is accepted in SIGKDD 2022.

Just to keep track of nice blog posts and new announcements related to machine learning, deep learning and artificial intelligence

Causal uplift modeling system for customer targeting and marketing ROI optimization

Code and frozen data from a pre-registered LLM-judge audit, with exploratory analyses of how rating-scale censoring can manufacture difference-in-differences effects. arXiv:2608.27309.

Clinical-AI Research Framework

Inference in Bayesian Belief Networks using Probability Propagation in Trees of Clusters (PPTC) and Gibbs sampling

Important link of cancer epidemiology and cancer prevention.

Causal ML pipeline for e-commerce dynamic pricing — Double Machine Learning for unbiased price elasticity, LightGBM demand forecasting (MAPE=0.418, R²=0.055), and a FastAPI pricing service delivering +30% revenue lift across 49,677 SKUs from 32M+ transactions.

Predicting Recession in Economy using various macro economic indicators

The Impact of Uber on Taxi Rides: A Causal Inference Study

Finds the invisible "second job" council employees do just navigating the organisation, chasing approvals, duplicate handovers, unclear ownership, and prices it in hours and FTE-days. Uses graph analysis, survival modelling and causal comparison, and never recommends cutting a job, only fixing the process.

A local-first RAG evaluation harness, and a worked example of an evaluation manufacturing its own result. Both effects the first version reported turned out to be artifacts of the judge rubric and a propensity score that was never estimated. All 16 trials and the audit are committed.

End-to-end SaaS churn, retention and revenue-risk analytics using Python, SQL, machine learning, calibration and causal uplift analysis.

R을 이용한 인과 추론 책.

Estimates whether an intervention actually caused an outcome, from observational data: propensity matching, IPW, S/T/X-learners, DiD and IV. Then tries to break its own result with refutation tests and an E-value — and reports "no effect" when that is the honest answer.

Statistical validity engine that audits A/B experiments across 8 checks:SRM, power, peeking, SUTVA, and more.

LLM-powered double-auction market simulation with Difference-in-Differences analysis and a Streamlit dashboard.

Uplift modeling & retention API — predicting which customers a retention contact can actually save. Live on Render