Jean-Michel Poggi

Variable selection using random forests

This paper proposes, focusing on random forests, the increasingly used statistical method for classification and regression problems introduced by Leo Breiman in 2001, to investigate two classical issues of variable selection. The first one is to find important variables for interpretation and the second one is more restrictive and try to design a good parsimonious prediction model. The main contribution is twofold: to provide some experimental insights about the behavior of the variable importance index based on random forests and to propose a strategy involving a ranking of explanatory variables using the random forests score of importance and a stepwise ascending variable introduction strategy.

Multivariate denoising using wavelets and principal component analysis

A multivariate extension of the well known wavelet denoising procedure widely examined for scalar valued signals, is proposed. It combines a straightforward multivariate generalization of a classical one and principal component analysis. This new procedure exhibits promising behavior on classical bench signals and the associated estimator is found to be near minimax in the one-dimensional sense, for Besov balls. The method is finally illustrated by an application to multichannel neural recordings.

Wavelets and their applications

Notations. Introduction. Chapter 1. A Guided Tour. Chapter 2. Mathematical Framework. Chapter 3. From Wavelet Bases to the Fast Algorithm. Chapter 4. Wavelet Families. Chapter 5. Finding and Designing a Wavelet. Chapter 6. A Short 1D Illustrated Handbook. Chapter 7. Signal Denoising and Compression. Chapter 8. Image Processing with Wavelets. Chapter 9. An Overview of Applications. Appendix: The EZW Algorithm. Bibliography. Index.

CLUSTERING FUNCTIONAL DATA USING WAVELETS

We present two methods for detecting patterns and clusters in high dimensional time-dependent functional data. Our methods are based on wavelet-based similarity measures, since wavelets are well suited for identifying highly discriminant local time and scale features. The multiresolution aspect of the wavelet transform provides a time-scale decomposition of the signals allowing to visualize and to cluster the functional data into homogeneous groups. For each input function, through its empirical orthogonal wavelet transform the first method uses the distribution of energy across scales generate a handy number of features that can be sufficient to still make the signals well distinguishable. Our new similarity measure combined with an efficient feature selection technique in the wavelet domain is then used within more or less classical clustering algorithms to effectively differentiate among high dimensional populations. The second method uses dissimilarity measures between the whole time-scale representations and are based on wavelet-coherence tools. The clustering is then performed using a k-centroid algorithm starting from these dissimilarities. Practical performance of these methods that jointly designs both the feature selection in the wavelet domain and the classification distance is demonstrated through simulations as well as daily profiles of the French electricity power demand.

Boosting and instability for regression trees

The AdaBoost like algorithm for boosting CART regression trees is considered. The boosting predictors sequence is analysed on various data sets and the behaviour of the algorithm is investigated. An instability index of a given estimation method with respect to some training sample is defined. Based on the bagging algorithm, this instability index is then extended to quantify the additional instability provided by the boosting process with respect to the bagging one. Finally, the ability of boosting to track outliers and to concentrate on hard observations is used to explore a non-standard regression context.

Clustering signals using wavelets

A wavelet-based procedure for clustering signals is proposed. It combines an individual signal preprocessing by wavelet denoising, a dimensionality reduction step by wavelet compression and a classical clustering strategy applied to a suitably chosen set of wavelet coefficients. The ability of wavelets to cope with signals of arbitrary or time-dependent regularity as well as to concentrate signal energy in few large coefficients, offers a useful tool to carry out both significant noise reduction and efficient compression. A simulated example and an electrical dataset are considered to illustrate the value of introducing wavelets for clustering such complex data.

A Test of Linearity for Functional Autoregressive Models

We propose a new test for linearity in time series. We consider an asymptotically stationary functional AR(p) model on ℜd of the form Xn = f(Xn−1, ..., Xn−p) + ξn (n∈ N). The testing procedure is based on a suitably normalized sum of quadratic deviations between two different estimates of the function f evaluated at q distinct points of ℜdp. The estimators are f^n, a recursive version of the non-parametric kernel estimator of f, and Ân, a least squares estimator well suited to the linear case. The main result states that the test statistic has a χ2 limit distribution under the null hypothesis. A similar result is derived under the alternative hypothesis for the test statistic corrupted by a non-linear term. Our simulations indicate that our asymptotic results hold for moderate sample sizes when the testing procedure is used carefully

TREND EXTRACTION FOR SEASONAL TIME SERIES USING ENSEMBLE EMPIRICAL MODE DECOMPOSITION

In this paper, we investigate eligibility of trend extraction through the empirical mode decomposition (EMD) and performance improvement of applying the ensemble EMD (EEMD) instead of the EMD for trend extraction from seasonal time series. The proposed method is an approach that can be applied on any time series with any time scales fluctuations. In order to evaluate our algorithm, experimental comparisons with three other trend extraction methods: EMD-energy-ratio approach, EEMD-energy-ratio approach, and the Hodrick–Prescott filter are conducted.

Nonlinear Adaptive Tracking Using Kernel Estimators: Estimation and Test for Linearity

We present some statistical results on nonlinear adaptive control using kernel estimators. We are concerned with a nonlinear autoregressive model of the form