CONCEPTUALIZATION OF NEURAL MODELS FOR DECISION SUPPORT TASKS

Abstract

This study delves into a comprehensive analysis of Artificial Neural Networks (ANNs) as tools for decision support and forecasting, juxtaposed with traditional models for statistically significant estimates and generalizations based on regression methods using client data series. Unlike conventional programming, ANNs offer a computing approach that does not require a complete algorithmic specification but is based on methods of inductive learning, automatic adaptation, and knowledge-based generalizations from big data. This makes them particularly useful for tasks where it might be difficult to formally define all the task parameters such as the size, description, and importance of input data, the expected number of layers and neurons per layer, activation functions, training parameters, regularization techniques, and expected losses and optimizations. These parameters are critical in decision-making under uncertainty, involving the analysis of large volumes of unstructured client data. ANNs employ an inductive method for collecting, storing, and utilizing expert knowledge, which is crucial for decision support systems. The literature review and summaries provided in this paper suggest that ANNs can be comparably effective to statistical models in forecasting and decision-making. The research highlights the potential contributions of ANNs to decision support systems and notes some limitations that could hinder achieving highly accurate results. It emphasizes the need to use numerous mathematical tools foundational to ANNs, such as convergence theorems, universal approximation theorems, optimization algorithms, and property analysis, to determine the best conditions for their application in specific forecasting tasks for recommendation systems and decision support systems. The study provides balanced assessments of the capabilities and limitations of ANNs in applied contexts such as oil and gas engineering, and in addressing issues related to supporting expert decision-making and predictive forecasting of operational and regulatory parameters.