Predictive Analytics and Causal Inferences

Background

Many application domains are faced with high-dimensional datasets where the researcher is either interested to build a predictive model, carry out the supervised or unsupervised classification of the available data from which statistical inference is performed. As a result, interpretable machine learning algorithms are becoming more popular and useful in solving real-world problems. While there are different approaches for the identification and selection of the best subset of variables to develop a predictive model, causal inference techniques have specifically become powerful tools to overcome the lack of interpretability for predictive analytics. Nonetheless, causal structure learning procedures are highly sensitive to the input, meaning the learned causal graph tends to change notably with data perturbations. Likewise, different causal learning algorithms find divergent causality conclusions. Therefore, developing robust frameworks poses a multifaceted challenge in preserving the maximum level of the information supplied by individual causal learners.

Objective

This research seeks to develop interpretable predictive analytics for high-dimensional and spatiotemporal datasets. The specific objectives are to: 1) tackle ill-conditioned covariance matrix estimates and yet develop explainable predictive models (vs. Blackbox AI models); and 2) develop robust ensemble frameworks for causality learning with heterogeneous graphs.  

Publication

Aslani, B., Mohebbi, S., (2023). “Ensemble framework for causality learning with heterogeneous Directed Acyclic Graphs through the lens of optimization”, Computers and Operations Research, 152, 106148.

Aslani, B., Mohebbi, S., Axthelm, H., (2021). “Predictive analytics for water main breaks using spatiotemporal data”, Urban water journal, 18(6), 433-448.

Mohebbi, S., Pamukcu, E., Bozdogan, H., (2019). “A new data adaptive elastic net predictive model using hybridized smoothed covariance estimators with information complexity”, Journal of Statistical Computation and Simulation, 89(6), 1060-1089.

Funding source

Faculty startup, George Mason University

GRA

TBD