Visual Clustering Approaches

Fig-1 The data stored in the world are rapidly growing. This growth of databases has far outpaced the human ability to interpret this data creating new phenomena of big data. Big data is difficult to work using most existing tools, automatic methods and visualization. Some new methods called Visual data Mining have recently appeared trying to involve more significantly the user in the data mining process and using more intensively the visualization. We think that is important to consider user perception to drawback dimension selection process, select initial seed of clustering algorithm, and select interactively the clusters for example. Inspired by these ideas, we propose a semi-interactive algorithm (Fig1 [1]) that we have developed, which integrate an automatic algorithm, an interactive evolutionary algorithm and visualization tools. This Approach can be applied on parsimonious clustering where clusters can be overlapping and we can detect interactively the relation between overlapped clusters ([2] Fig 2). We propose also an interactive clustering approach, where we can identify visually clusters by combining different projections of interactive visualizations. In many situations where visual perception is more effective than classical clustering methods, the proposed approach gives better results ([3] Fig 3). We can apply these techniques combined with a new iterative clustering approach that extract compact clusters one by one, in this aspect the visualization is very important, the user can stop or continue the process according to the obtained information [4]. These approaches can be applied on different problems on big data: data steam, social networks [5], …