Search

International audience

International audience; —The field of time series forecasting has progressed significantly in recent decades, specially in regards to the need of forecasting economic data. That said, some issues still arise. In particular when we are working with a set of time series that have a large number of variables. Hence, a selection step is usually needed in order to reduce the number of variables that will contribute to forecast each target time series. In this paper, we propose a feature selection and / or dimension reduction algorithm for forecasting multivariate time series, based on (i) the notion of the Granger causality, and (ii) on a selection step based on a clustering strategy. Finally, we carry out experiments on different real data sets, by comparing our proposal and some of the most used feature selection methods. Experiments show that we improved the forecasting accuracy compared with the evaluated methods.

International audience; Handling time series forecasting with many predictors is a popular topic in the era of "Big data", where wast amounts of observed variables are stored and used in analytic processes. Classical prediction models face some limitations when applied to large-scale data. Using all the existing predictors increases the computational time and does not necessarily improve the forecast accuracy. The challenge is to extract the most relevant predictors contributing to the forecast of each target time series. We propose a causal-feature selection algorithm specific to multiple time series forecasting based on a clustering approach. Experiments are conducted on US and Australia macroeconomic datasets using different prediction models. We compare our method to some widely used dimension reduction and feature selection methods including principal component analysis PCA, Kernel PCA and factor analysis. The proposed algorithm improves the forecast accuracy compared to the evaluated methods on the tested datasets.

16 pages; International audience; In this paper, we present a new approach relevant to the discovery of correlated patterns, based on the use of multicore architectures. Our work rests on a full KDD system and allows one to extract Decision Correlation Rules based on the Chi-squared criterion that include a target column from any database. To achieve this objective, we use a levelwise algorithm as well as contingency vectors, an alternate and more powerful representation of contingency tables, in order to prune the search space. The goal is to parallelize the processing associated with the extraction of relevant rules. The parallelization invokes the PPL (Parallel Patterns Library), which allows a simultaneous access to the whole available cores / processors on modern computers. We nally present rst results on the reached performance gains.

6 pages; International audience; In this paper, we present a new approach relevant to the discovery of correlated patterns, based on the use of multicore architectures. Our work rests on a full KDD system and allows one to extract Decision Correlation Rules based on the Chi-squared criterion that include a target column from any database. To achieve this objective, we use a levelwise algorithm as well as contingency vectors, an alternate and more powerful representation of contingency tables, in order to prune the search space. The goal is to parallelize the processing associated with the extraction of relevant rules. The parallelization invokes the PPL (Parallel Patterns Library), which allows a simultaneous access to the whole available cores / processors on modern computers. We finally present first results on the reached performance gains.

7 pages; International audience; Yield enhancement is a key issue in semiconductor manufacturing. Data mining tools can therefore be helpful, by extracting hidden links between numerous complex process control parameters. In order to highlight correlations between such parameters, we developed a complete Knowledge Discovery in Databases (KDD) model, called MineCor. Its mining heart uses a new method derived from association rules programming, based on lectic search and contingency vectors. After recalling these concepts, this paper focuses on data preprocessing and transformation functions, which have an important impact on final results. An overall presentation of these functions, of some significant experimental results and of associated performances are provided and finally discussed.

6 pages; International audience; Nowadays, data mining tools are becoming more and more popular to extract knowledge from a huge volume of data. In this paper, our aim is to extract Literal Correlation Rules: Correlation Rules admitting literal patterns given a set of items and a binary relation. If a pattern represents a valid Correlation Rule, then any literal belonging to its Canonical Base represents a valid Literal Correlation Rule. Moreover, in order to highlight only relevant Literal Correlation Rules, we add a pruning step based on a support threshold. To extract such rules, we modify the LHS-CHI2 Algorithm and perform some experiments.

Actes des neuvièmes Journées Extractions et gestion des connaissances EGC'2009, Strasbourg du 27 au 30 janvier 2009. - Numéro de : "Revue des nouvelles technologies de l'information", ISSN 1764-1667, RNTI-E-15, 2009; International audience; Dans cet article, nous introduisons deux nouveaux concepts : les règles de corrélation décisionnelles et les vecteurs de contingence. Le premier résulte d'un couplage entre les règles de corrélation et les règles de décision. Il permet de mettre en évidence des liens pertinents entre certains ensembles de motifs d'une relation binaire et les valeurs d'un attribut cible (appartenant à cette même relation) en se basant à la fois sur la mesure du Khi-carré et sur le support des motifs extraits. De par la nature du problème, les algorithmes par niveaux font que l'extraction des résultats a lieu avec des temps de réponse élevés et une occupation mémoire importante. Afin de palier à ces deux inconvénients, nous proposons un algorithme basé sur l'ordre lectique et les vecteurs de contingence.

National audience