OPTIMIZATION OF INDEXING IN DATABASE MANAGEMENT SYSTEMS BASED ON ADAPTIVE DEEP LEARNING MODELS

Abstract

This research is devoted to studying the problem of improving indexing efficiency in database management systems (DBMS) using deep learning algorithms. Indexing remains a key aspect that affects query performance, data search efficiency, and the overall response speed of DBMS operations. Traditional methods of creating and optimising indexes are mainly rule-based and require significant manual configuration. Such approaches are inherently non-adaptive, which creates significant limitations in dynamic environments characterised by load fluctuations and changing data models.

The research focuses primarily on developing an improved SQL query classification model aimed at improving the automation of indexing processes. Using deep learning techniques and Transformer-based neural network architectures, the proposed model is capable of independently recognising query patterns and predicting optimal indexing strategies, thus eliminating the need for human intervention. This study uses methodologies based on big data analysis, numerical modelling, and experimental validation, using the open Spider dataset, which covers a wide range of diverse SQL query structures obtained from different domains.

The experimental results demonstrate that the implemented model significantly reduces both the average time of execution of SQL queries and the processing resources required for index generation.. Compared to traditional approaches, the deep learning-based solution demonstrates better scalability, adaptability, and accuracy in recognising dependencies related to indexing in complex queries.

The suggested method is of important practical interest for modern information systems, especially in such fields as cloud computing, financial data processing, and high-volume distributed infrastructures. Allowing database administrators and automated systems to dynamically improve indexing strategies, the model enables stable and reliable system operation. Future research will focus on integrating reinforcement learning methods to improve the model's capabilities and evaluating its suitability for use in distributed DBMS architectures to optimise query processing and resource management.