VISDMiner: An Interactive Data Mining Process Visualization System

Yong-sheng WANG,Hui LI,Mei CHEN,Zhen-yu DAI,Ming ZHU

DOI: https://doi.org/10.3969/j.issn.1006-2475.2018.06.014

2018-01-01

Abstract:In order to address the problem that the data mining process often to be not transparent and lack of user interaction, we design and implement the VISDMiner system. VISDMiner combines the visualization technology and the data mining technology to provide the capability of analyzing and visualizing partial results of all stages of mining process. During the procedures, users can tune the parameters of data mining algorithm and visualizations according to their domain knowledge and experience to achieve fur-ther data exploration. In order to deal with high-dimensional data set, VISDMiner system uses an improved algorithm MIC-PCA for principal component analysis based on the maximum information coefficient. The algorithm is mainly aimed at improving the dimensionality reduction and classification accuracy of traditional PCA algorithms. The experimental results show that VISDMine not only realizes the visualization of the data mining process, but also improves the user's understandability of the data mining re-sults,and the MIC-PCA algorithm also improves the dimensionality reduction and classification accuracy of PCA algorithm.