Performance Improvement Algorithms in Big Data Analysis

Abstract

Abstract This article describes the study results of development methods and algorithms for performance improvement in the big data analysis and training of machine learning models. The data growing and complexity problem describing complex processes require new approaches and tools for the scientific and technical community. In the course of research, the algorithms for analysis of heterogeneous medical and law records were performed. The performance improvement in classification, clustering and graph calculation problems were solved. With using CUDA it was possible to get more than 95 times performance. The usage of high-performance technologies is important in the analysis of electronic records because it provides an adequate response to the process of analysis of large scale of data from information systems. This study shows how to speed up the calculations on the example of most basic and widespread machine learning tasks. The results of the study can be used to develop a new generation of decision support systems, interactive data analysis systems and methods for eScience