Interactive visualization for knowledge discovery
| dc.contributor.author | Han, Jianchao | en |
| dc.date.accessioned | 2006-07-28T19:11:18Z | |
| dc.date.available | 2006-07-28T19:11:18Z | |
| dc.date.issued | 2001 | en |
| dc.date.submitted | 2001 | en |
| dc.description.abstract | Knowledge discovery in databases (KDD) has been actively pursued. Currently, two approaches are extensively researched and developed. One is algorithm-based and the other is visualization-based. The algorithm-based methods specify the target formats and pursue the correlations between the outcome variable and the independent variables. Once the algorithms are determined, the user can hardly participate in the discovery process. The visualization-based methods specify the hypothesis by means of visualization metaphors and pursue interactive visualization of large data sets and the patterns behind the data sets. Most visualization systems lack the ability to visualize the entire process of knowledge discovery, neither consider to include the users' perception into the systems. We propose an interactive visualization model, RuleViz, for the KDD process, which stresses the human-machine interaction and visual representation. The interaction between the user and the machine helps the KDD system navigate throuhg the enormous search spaces and recognize the intentions of the user. Thus, the user can easily provide the system with heuristics and domain knowledge and specify parameters. On the other hand, the visual representation of data and knowledge resulted in the KDD process helps users gain better insight into multidimensional data, understand the intermediate results, and interpret the discovered patterns. RuleViz consists of five components: raw data preparation and visualization, interactive data reduction (jorizontally and vertically), visual data preprocessing such as missing values handling, numerical attribute discretization and data transformation, pattern discovery like correlation mining, rule induction and decision tree construction, and pattern visualization like neural networks, decision trees, and classification list. To implement the RuleViz model, we suggest three implementation paradigms: image-based algorithmic implementation, embedded-algorithm-based implementation, and user-supervised interactive implementation, and implement four interactive knowledge discovery systems: AViz - an image-based system for discovering numerical association rules based on data plots and optimized rectangles; CViz - an embedded-algorithm-based system for classification rule induction based on the parallel coordinates visualization technique: CVizT - a user-supervised interactive system for building classification rules based on the Table Lens visualization technique; and DTViz - a user-supervised interactive system for constructing decision trees based on the parallel segments pixel-oriented visualization technique and the tree structure visualization algorithm. Our experimental results with the UCI repository data sets and artificial data sers demonstrate that the RuleViz model can provide a methodology for developing interactive KDD systems. The systems developed according to the RuleViz model take advantage of both algorithm-based and visualization-based approaches. They provide the user with straightforward observation and control of the KDD process, integrate the user's perception and domain knowledge into the KDD process easily, make it convenient for the user to understand and interpret the discovered knowledge, and remain flexible for distributing the KDD functions between the user and machine. | en |
| dc.format | application/pdf | en |
| dc.format.extent | 15966919 bytes | |
| dc.format.mimetype | application/pdf | |
| dc.identifier.uri | http://hdl.handle.net/10012/621 | |
| dc.language.iso | en | en |
| dc.pending | false | en |
| dc.publisher | University of Waterloo | en |
| dc.rights | Copyright: 2001, Han, Jianchao. All rights reserved. | en |
| dc.subject | Harvested from Collections Canada | en |
| dc.title | Interactive visualization for knowledge discovery | en |
| dc.type | Doctoral Thesis | en |
| uws-etd.degree | Ph.D. | en |
| uws.peerReviewStatus | Unreviewed | en |
| uws.scholarLevel | Graduate | en |
| uws.typeOfResource | Text | en |
Files
Original bundle
1 - 1 of 1