Adaptive Computational Models for Intelligent Data-Driven Decision Systems in Complex Environments

Abstract

Increasingly, modern decision systems are being executed in complex, uncertain and dynamic environments, ranging with autonomous vehicles and smart grids to adaptive healthcare monitoring. The paper reviews and summarizes adaptive computational frameworks that can support robust and data-driven decision making in these circumstances. An architectural taxonomy (learn, infer, adapt, and control layers) is defined, methodological decisions (probabilistic modeling, reinforcement learning, fuzzy logic, ensemble and hybrid models) described and a worked example where we perform statistical analysis to show how adaptation to nonstationary data can be done. The discussion outlines the main trade-offs (consistency vs. explanativeness, sample effectiveness vs. versatility), the presence of the in-text citations to the underlying literature, and the recommendations on evaluation and implementation. We provide tables, figures (descriptive diagrams), pseudocode and a plan of statistical analysis that can be reproduced. At the end of the paper, research and practice suggestions are provided.