[1Yin-B] Poster 2

[1Yin-B-60] Individual Case Analysis Method using Rule-Based Probabilistic Prediction Models and Counterfactual Simulation

〇Akari Oda¹, Yoichi Seki¹, Kaoru Shimada¹ (1. Univ of Gunma)

Keywords:

Association Rule,Logistic regression analysis,instance-specific model

This study proposes a method that integrates interpretable association rules with a logistic function to generate probabilistic predictions for classification problems. A counterfactual simulation framework is introduced to analyze predicted probabilities and their changes by virtually modifying input attributes. ItemSB (Itemsets with Statistically Distinctive Backgrounds) are discovered for each instance using the evolutionary computation method GNMiner. By embedding a logistic curve in the consequent of each rule, continuous probability outputs are obtained, extending conventional discrete rule representations to instance-level probabilistic prediction. Based on the constructed instance-specific models, attributes associated with probability increases are identified and used as explanatory variables in instance-wise logistic regression analyses. Evaluating probability changes under hypothetical attribute modifications enables interpretable analysis of the relationship between input attributes and prediction outcomes for each instance.