Abstract

Retail organizations increasingly require analytical systems that can move beyond prediction toward explanations of observed business changes and structured decision support. This paper presents ADIF, an end-to-end decision intelligence framework that connects predictive modeling, predictive-driver discovery, causal analysis, counterfactual reasoning, and large language model (LLM)-assisted communication in a single workflow. The framework combines XGBoost and SHAP for predictive modeling and feature attribution, the PC algorithm for causal-structure screening, DoWhy for treatment-effect estimation and refutation, counterfactual analysis for scenario evaluation, and Llama 3.1 through Ollama for natural-language decision support. ADIF was evaluated on the Online Retail II and Instacart datasets. On Online Retail II, Order Change and Quantity Change represented 63.42% and 36.58%, respectively, of the reported normalized feature-attribution contribution. The estimated effect of Quantity Change on Revenue Change was 0.59748, with a reported 95% confidence interval of approximately [0.5721, 0.6229]. On Instacart, the estimated effect of Orders on Reorder Rate was 0.006417, with a reported 95% confidence interval of approximately [0.00591, 0.00693]. Predictive evaluation produced a mean cross-validated R2 of 0.5883 for XGBoost on Instacart. The counterfactual stage was configured for 5%, 10%, 20%, and 50% treatment-change scenarios as model-based what-if analyses; no scenario outcome is claimed as a verified finding. Refutation experiments produced a negligible change under the random-common-cause test and a near-zero estimate under placebo treatment; these are treated as sensitivity checks rather than proof of causality. The LLM layer converts structured quantitative findings and limitations into decision-support narratives. The framework’s contribution is the explicit evidence handoff between analytical stages while keeping predictive attribution, causal estimation, counterfactual scenarios, and language generation conceptually separate.

References

• T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proc. 22nd ACM SIGKDD Int. Conf.
• Knowledge Discovery and Data Mining, 2016, pp. 785–794.
• L. Breiman, “Random Forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001.
• G. Shmueli, “To Explain or to Predict?” Statistical Science, vol. 25, no. 3, pp. 289–310, 2010.
• S. M. Lundberg and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” in Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 4765–4774.
• P. Spirtes, C. Glymour, and R. Scheines, Causation, Prediction, and Search, 2nd ed., MIT Press, 2000.
• J. Pearl, Causality: Models, Reasoning, and Inference, 2nd ed., Cambridge University Press, 2009.
• D. B. Rubin, “Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies,” Journal of Educational Psychology, vol. 66, no. 5, pp. 688–701, 1974.
• G. W. Imbens, “Nonparametric Estimation of Average Treatment Effects under Exogeneity: A Review,” Review of Economics and Statistics, vol. 86, no. 1, pp. 4–29, 2004.
• A. Sharma and E. Kiciman, “DoWhy: An End-to-End Library for Causal Inference,” arXiv preprint arXiv:2011.04216, 2020.
• S. Wachter, B. Mittelstadt, and C. Russell, “Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR,” Harvard Journal of Law & Technology, vol. 31, no. 2, pp. 841–887, 2018.
• A. Grattafiori et al., “The Llama 3 Herd of Models,” arXiv preprint arXiv:2407.21783, 2024.
• D. Chen, “Online Retail II,” UCI Machine Learning Repository, 2019, doi: 10.24432/C5CG6D.
• Instacart, “The Instacart Online Grocery Shopping Dataset 2017,” 2017. Available: https://www.instacart.com/ datasets/grocery-shopping-2017.