Alain Dussauchoy

Parameter estimation of the generalized gamma distribution

This article focuses on the parameter estimation of the generalized gamma distribution. Because of many difficulties described in the literature to estimate the parameters, we propose here a new estimation method. The algorithm associated to this heuristic method is implemented in Splus. We validate the resulting routine on the particular cases of the generalized gamma distribution.

A new clustering approach for symbolic data and its validation: application to the healthcare data

Graph coloring is used to characterize some properties of graphs. A b-coloring of a graph G (using colors 1,2,...,k) is a coloring of the vertices of G such that (i) two neighbors have different colors (proper coloring) and (ii) for each color class there exists a dominating vertex which is adjacent to all other k-1 color classes. In this paper, based on a b-coloring of a graph, we propose a new clustering technique. Additionally, we provide a cluster validation algorithm. This algorithm aims at finding the optimal number of clusters by evaluating the property of color dominating vertex. We adopt this clustering technique for discovering a new typology of hospital stays in the French healthcare system.

A Graph b-coloring Framework for Data Clustering

The graph b-coloring is an interesting technique that can be applied to various domains. The proper b-coloring problem is the assignment of colors (classes) to the vertices of one graph so that no two adjacent vertices have the same color, and for each color class there exists at least one dominating vertex which is adjacent (dissimilar) to all other color classes. This paper presents a new graph b-coloring framework for clustering heterogeneous objects into groups. A number of cluster validity indices are also reviewed. Such indices can be used for automatically determining the optimal partition. The proposed approach has interesting properties and gives good results on benchmark data set as well as on real medical data set.

Clinical pathway analysis using graph-based approach and Markov models

Cluster analysis is one of the most important aspects in the data mining process for discovering groups and identifying interesting distributions or patterns over the considered data sets. A new method for sequences clustering and prediction is presented in this paper, which is based on a hybrid model that uses our b-coloring based clustering approach as well as Markov chain models. The paper focuses on clinical pathway analysis but the method applies to every kind of sequences, and a generic decision support framework has been developed for managers and experts. The interesting result is that the clusters obtained have a twofold representation. Firstly, there is a set of dominant sequences which reflects the properties of the cluster and also guarantees that clusters are well separated within the partition. On the other hand, the behavior of each cluster is governed by a finite-state Markov chain model which allows probabilistic prediction. These models can be used for predicting possible paths for a new patient, and for helping medical professionals to eventually react to exceptions during the clinical process.

Generalized Extreme Value distribution for fitting opening/closing asset prices and returns in stock- exchange

Robust estimation of stock-exchange fluctuations is a challenging problem. The accuracy of statistical extrapolation is fairly sensitive to both model and sampling error. Using the opening/closing quotation and return data (concerning stock-exchange), this paper presents a comparative assessment using various theoretical distributions: Normal, LogNormal, Gamma, Gumbel, Weibull, Generalized Extreme Value (GEV).We used GEV distribution in an other context than extreme value theory (indeed dedicated to this domain). From the empirical distribution on short periods (3, 6, 9 and 12 months), we prove that GEV distribution allows to correctly fit returns and opening/closing quotations (without studying only the behaviour of maxima or minima in a sample, but overall of the sample) by comparison with the other distributions. This paper focuses on the GEV distribution in the univariate case. Following a review of the literature, univariate GEV distribution is applied to a series of daily stock-exchange of TOTAL oil company. We illustrate this article with the opening/closing quotations minus the moving average of the five last days and the returns of this company on short and medium terms (3, 6, 9, 12 months moving forward 1 month).

A partially dynamic clustering algorithm for data insertion and removal

We consider the problem of dynamic clustering which has been addressed in many contexts and applications including dynamic information retrieval, Web documents classification, etc. The goal is to efficiently maintain homogenous and well-separated clusters as new data are inserted or existing data are removed. We propose a framework called dynamic b-coloring clustering based solely on pairwise dissimilarities among all pairs of data and on cluster dominance. In experiments on benchmark data sets, we show improvements in the performance of clustering solution in terms of quality and computational complexity.

A new greedy algorithm for improving b-coloring clustering

This paper proposes a new greedy algorithm to improve the specified b-coloring partition while satisfying b-coloring property. The b-coloring based clustering method in [3] enables to build a fine partition of the data set (classical or symbolic) into clusters even when the number of clusters is not pre-defined. It has several desirable clustering properties: utilization of topological relations between objects, robustness to outliers, all types of data can be accommodated, and identification of each cluster by at least one dominant object. However, it does not consider the high quality of the clusters in the construction of a b-coloring graph. The proposed algorithm in this paper can complement its weakness by re-coloring the objects to improve the quality of the constructed partition under the property and the dominance constraints. The proposed algorithm is evaluated against benchmark datasets and its effectiveness is confirmed.

Constrained graph b-coloring based clustering approach

Clustering is generally defined as an unsupervised data mining process which aims to divide a set of data into groups, or clusters, such that the data within the same group are similar to each other while data from different groups are dissimilar. However, additional background information (namely constraints) are available in some domains and must be considered in the clustering solutions. Recently, we have developed a new graph b-coloring clustering algorithm. It exhibits more important clustering features and enables to build a fine partition of the data set in clusters when the number of clusters is not pre-defined. In this paper, we propose an extension of this method to incorporate two types of Instance-Level clustering constraints (must-link and cannot-link constraints). In experiments with artificial constraints on benchmark data sets, we show improvements in the quality of the clustering solution and the computational complexity of the algorithm.

A New Graph-Based Clustering Approach: Application to PMSI Data

Graph coloring is used to characterize some properties of graphs. A b-coloring of a graph G (using colors 1,2,...,k) is a coloring of the vertices of G such that (i) two neighbors have different colors (proper coloring) and (ii) for each color class there exists a dominating vertex which is adjacent to all other k-1 color classes. In the French healthcare system, the classification of patients into diagnosis related groups (DRGs) is performed using a supervised approach according to a decision tree. The main problem of this classification scheme concerns the heterogeneity of several DRGs resulting from the variety of pathology and examinations within the DRG class. In this paper, we propose a new approach of clustering based on a b-coloring of graphs to define a typology of patients

A Knowledge Management Based Framework as a Way for SME Networks Integration

This paper initially introduces specificities of small and medium enterprises (SME) networks, then deals with the recent evolution of the Knowledge Management (KM). In coherence with this evolution, we describe a knowledge management process based on a community of practices which can be applied to these types of groups. The main conclusion is that knowledge management projects provide a good manner for SME networks to facilitate and increase the collaboration rate as well as to share knowledge, allowing them to make collaborative work more efficient.