2016/05/22 by Ridho Rahmadi, Rahmadi, Ridho, Groot, Perry +10
Computer Science · Decision Sciences · Mathematics · #Advanced Causal Inference Techniques #Artificial Intelligence (cs.AI) #Bayesian Modeling and Causal Inference #FOS: Computer and information sciences #Machine Learning (stat.ML) #Multi-Criteria Decision Making
paper · pdf · doi:10.48550/arxiv.1605.06838
openalex publication_date 2016/05/22 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
A typical problem in causal modeling is the instability of model structure\nlearning, i.e., small changes in finite data can result in completely different\noptimal models. The present work introduces a novel causal modeling algorithm\nfor longitudinal data, that is robust for finite samples based on recent\nadvances in stability selection using subsampling and selection algorithms. Our\napproach uses exploratory search but allows incorporation of prior knowledge,≠.g., the absence of a particular causal relationship between two specific\nvariables. We represent causal relationships using structural equation models.\nModels are scored along two objectives: the model fit and the model complexity.\nSince both objectives are often conflicting we apply a multi-objective\nevolutionary algorithm to search for Pareto optimal models. To handle the\ninstability of small finite data samples, we repeatedly subsample the data and\nselect those substructures (from the optimal models) that are both stable and\nparsimonious. These substructures can be visualized through a causal graph. Our\nmore exploratory approach achieves at least comparable performance as, but\noften a significant improvement over state-of-the-art alternative approaches on\na simulated data set with a known ground truth. We also present the results of\nour method on three real-world longitudinal data sets on chronic fatigue\nsyndrome, Alzheimer disease, and chronic kidney disease. The findings obtained\nwith our approach are generally in line with results from more\nhypothesis-driven analyses in earlier studies and suggest some novel\nrelationships that deserve further research.\n