MAIN METHODS OF MACHINE LEARNING IN DECISION SUPPORT SYSTEMS

Abstract

The purpose of this paper is a comprehensive review and systematization of the main classes of machine learning (ML) methods utilized in decision support systems (DSS). The research includes a detailed analysis of distinctive features, practical examples of ML algorithms application in various fields (medicine, finance, logistics), and a critical comparison of the advantages and disadvantages of each method to identify optimal conditions for their integration into DSS and to enhance decision-making efficiency. The research methodology is based on a comprehensive analysis of scientific publications devoted to the use of ML in DSS, employing classification methods of algorithms according to the type of learning (supervised, unsupervised, semisupervised, reinforcement learning, deep learning). The study utilizes methods of analysis, synthesis, comparative evaluation of algorithm capabilities, and generalization of practical implementation results. Additionally, the research considers the quality of input data used for training models and addresses potential ethical and technological constraints related to the integration of these methods. The scientific novelty of this research lies in the first-time holistic and systematized assessment of various categories of ML methods within DSS, providing an in-depth analysis of their advantages, limitations, the most suitable areas of application, and the conditions determining their integration effectiveness. The paper generalizes and compares the outcomes of previous studies, forming a comprehensive vision of prospective directions for the development of intelligent DSS, taking into account contemporary technological and ethical challenges. The conclusions of the research confirm that the integration of ML methods significantly enhances the quality and validity of managerial decisions by automating the analysis of large data volumes and adapting recommendations based on accumulated experience. It is established that selecting a specific class of methods should depend on the nature of the task, the volume of available labeled data, the requirements for model interpretability, and allowable resource expenditures. Further research perspectives involve developing methodological guidelines for the optimal selection of methods for various types of DSS, improving algorithm transparency and interpretability, and creating effective mechanisms to ensure their ethical use, reliability, and safety in practical applications.