François Poulet

SVM and graphical algorithms: a cooperative approach

We present a cooperative approach using both support vector machine (SVM) algorithms and visualization methods. SVM are widely used today and often give high quality results, but they are used as black-box (it is very difficult to explain the obtained results) and cannot treat easily very large datasets. We have developed graphical methods to help the user to evaluate and explain the SVM results. The first method is a graphical representation of the separating frontier quality, it is then linked with other visualization tools to help the user explaining SVM results. The information provided by these graphical methods is also used for SVM parameter tuning, they are then used together with automatic algorithms to deal with very large datasets on standard computers. We present an evaluation of our approach with the UCI and the Kent Ridge Bio-medical data sets.

Visual interactive evolutionary algorithm for high dimensional data clustering and outlier detection

Usual visualization techniques for multidimensional data sets, such as parallel coordinates and scatter-plot matrices, do not scale well to high numbers of dimensions. A common approach to solve this problem is dimensionality selection. Existing dimensionality selection techniques usually select pertinent dimension subsets that are significant to the user without loose of information. We present concrete cooperation between automatic algorithms, interactive algorithms and visualization tools: the evolutionary algorithm is used to obtain optimal dimension subsets which represent the original data set without loosing information for unsupervised mode (clustering or outlier detection). The last effective cooperation is a visualization tool used to present the user interactive evolutionary algorithm results and let him actively participate in evolutionary algorithm searching with more efficiency resulting in a faster evolutionary algorithm convergence. We have implemented our approach and applied it to real data set to confirm this approach is effective for supporting the user in the exploration of high dimensional data sets.

MINING VERY LARGE DATASETS WITH SUPPORT VECTOR MACHINE ALGORITHMS

In this paper, we present new support vector machines (SVM) algorithms that can be used to classify very large datasets on standard personal computers. The algorithms have been extended from three recent SVMs algorithms: least squares SVM classification, finite Newton method for classification and incremental proximal SVM classification. The extension consists in building incremental, parallel and distributed SVMs for classification. Our three new algorithms are very fast and can handle very large datasets. An example of the effectiveness of these new algorithms is given with the classification into two classes of one billion points in 10-dimensional input space in some minutes on ten personal computers (800 MHz Pentium III, 256 MB RAM, Linux).

Enhancing SVM with Visualization

Understanding the result produced by a data-mining algorithm is as important as the accuracy. Unfortunately, support vector machine (SVM) algorithms provide only the support vectors used as “black box” to efficiently classify the data with a good accuracy. This paper presents a cooperative approach using SVM algorithms and visualization methods to gain insight into a model construction task with SVM algorithms. We show how the user can interactively use cooperative tools to support the construction of SVM models and interpret them. A pre-processing step is also used for dealing with large datasets. The experimental results on Delve, Statlog, UCI and bio-medical datasets show that our cooperative tool is comparable to the automatic LibSVM algorithm, but the user has a better understanding of the obtained model.

Kernel-based Algorithms and Visualization for Interval Data Mining

Our investigation aims at extending kernel methods to interval data mining and using graphical methods to explain the obtained results. Interval data type can be a good way to aggregate large datasets into smaller ones or to represent data with uncertainty. No algorithmic changes are required from the usual case of continuous data other than the modification of the radial basis kernel function evaluation. Thus, kernel-based algorithms can deal easily with interval data. The numerical test results with real and artificial datasets show that the proposed methods have given promising performance. We also use interactive graphical decision tree algorithms and visualization techniques to give an insight into support vector machines results. The user has a better understanding of the models behaviour

High dimensional visual data classification

We present new visual data mining algorithms for interactive decision tree construction with large datasets. The size of data stored in the world is constantly increasing but the limits of current visual data mining (and visualization) methods concerning the number of items and dimensions of the dataset treated are well known (even with pixellisation methods). One solution to improve these methods is to use a higher-level representation of the data, for example a symbolic data representation. Our new interactive decision tree construction algorithms deal with interval and taxonomical data. With such a representation, we are able to deal with potentially very large datasets because we do not use the original data but higher-level data representation. Interactive algorithms are examples of new data mining approach aiming at involving more intensively the user in the process. The main advantages of this user-centered approach are the increased confidence and comprehensibility of the obtained model, because the user was involved in its construction and the possible use of human pattern recognition capabilities. We present some results we obtained on very large datasets.

User guidance: from theory to practice, the case of visual data mining

We present two modules for users guidance in visual data mining (VDM) domain which use artificial intelligence. User guidance refers to the available ways to advise, orient, instruct, and inform the users throughout their interactions with a computer. The first module allows recommending best available data mining algorithms to users and the second module aims at reducing attributes and/or data items in very large data sets to be visually mined

Théorie du consensus appliquée au prétraitement des ensembles de données

Visualization methods do not scale well with high number of features. We present an approach using a consensus theory based feature selection (CTBFS) algorithm, clustering for sampling and visualization for weight assignment in order to aggregate multivariate and multidimensional datasets. We use datasets available in UCI and Kent Ridge Bio Medical Dataset Repositories in order to evaluate the performance of our new approach.