Research On High-dimensional Data Visualization And Application In Business Intelligence

Technology of High-Dimensional Data Visualization has become an important andeffective way to obtain useful information from high-dimensional data, which is mainlyused for high-dimensional data analysis and presentation. HDDV is playing an increasinglyimportant role in some fields, such as image recognition, biological data, earthquakeprediction, space remote sensing, etc. With the development of HDDV technology, how tograsp the useful information in high-dimensional data and further through visualizationtechnology to achieve low-dimensional display, which is an urgent need to address theproblem in the present. This firstly needs necessary processing by virtue of dimensionreduction method for high-dimensional data, and then by using the visualization technologyto realize eventual display for graphical analysis.Due to the characteristics of high dimensional data itself, on the one hand, whichcauses the time and space complexity very high as well as makes us in a dilemma, on theother hand, it also contains a lot of information people can’t be obtained directly through theobservation. For handling high-dimensional data, a lot of work is focused on the study ofdimension reduction algorithms, especially manifold learning. We can use geometry-basedvisualization techniques in order to realize the display of high-dimensional data, such asparallel coordinates method. For high-dimensional data visualization, this paper mainlyconcentrates on the following aspects:1. This paper conducts some research related to the linear and nonlinear dimensionalityreduction method. And proposes some improvements contrary to dimension reductionmethod of manifold learning, mainly studying some aspects, such as, effectively extractintrinsic dimension of high-dimensional data, improving the classification efficiency,achieving incremental learning, reducing noise interference, etc. Meanwhile, this papergives the corresponding experimental demonstration.2. For visualization technology of parallel coordinates, we can consider limitingdimensions or reducing the dimension in order to make up for the lack of visual effects dueto intensive data.3. For data mining process of business intelligence, we can add a preliminaryvisualization process, and this paper gives preliminary visualization method of based ondimension restrictions and Pareallel Coordinates based on Isomap and MLE, in order toprovide a good feedback to ultimate data mining. Experimental results demonstrate that improved dimension reduction algorithms basedon a standard set of data is indeed able to compensate for the lack of dimensionalityreduction methods in some aspects, which include Isomap, LLE and LTSA. In this paper, wetake two kinds of data in business intelligence as an example, and give the detailedimplementation process of preliminary visualization, which achieves good results.