Héctor Allende-Cid

Self-Organizing Neuro-Fuzzy Inference System

The architectural design of neuro-fuzzy models is one of the major concern in many important applications. In this work we propose an extension to Rogerss ANFIS model by providing it with a selforganizing mechanism. The main purpose of this mechanism is to adapt the architecture during the training process by identifying the optimal number of premises and consequents needed to satisfy a users performance criterion. Using both synthetic and real data, our proposal yields remarkable results compared to the classical ANFIS.

Robust alternating AdaBoost

Ensemble methods are general techniques to improve the accuracy of any given learning algorithm. Boosting is a learning algorithm that builds the classifier ensembles incrementally. In this work we propose an improvement of the classical and inverse AdaBoost algorithms to deal with the problem of the presence of outliers in the data. We propose the Robust Alternating AdaBoost (RADA) algorithm that alternates between the classic and inverse AdaBoost to create a more stable algorithm. The RADA algorithm bounds the influence of the outliers to the empirical distribution, it detects and diminishes the empirical probability of bad samples, and it performs a more accurate classification under contaminated data. n nWe report the performance results using synthetic and real datasets, the latter obtained from a benchmark site.

Context-Aware Regression from Distributed Sources

In this paper we present a distributed regression framework to model data with different contexts. Different context is defined as the change of the underlying laws of probability in the distributed sources. Most state of the art methods do not take into account the different context and assume that the data comes from the same statistical distribution. We propose an aggregation scheme for models that are in the same neighborhood in terms of statistical divergence.We conduct experiments with synthetic data sets to validate our proposal. Our proposed algorithm outperforms other models that follow a traditional approach.

A Flexible Neuro-Fuzzy Autoregressive Technique for Non-linear Time Series Forecasting

The aim of this paper is to simultaneously identify and estimate a non-linear autoregressive time series using a flexible neuro-fuzzy model. We provide a self organization and incremental mechanism to the adaptation process of the neuro-fuzzy model. The self organization mechanism searches for a suitable set of premises and consequents to enhance the time series estimation performance, while the incremental method selects influential lags in the model description. n nExperimental results indicate that our proposal reliably identifies appropriate lags for non-linear time series. Our proposal is illustrated by simulations on both synthetic and real data.

Neuro-fuzzy-based Arrhythmia Classification Using Heart Rate Variability Features

Arrhythmia diagnosis is commonly conducted through visual analysis of human electrocardiograms, a very resource consuming task for physicians. In this paper we present a computational approach for arrhythmia detection based on heart rate variability signal analysis and the application of a neuro-fuzzy classification model called SONFIS. The aforementioned method generates a set of linguistically interpretable inference rules for pattern classification and outperforms artificial neural networks and support vector machines in accuracy and several other performance indicators.

SONFIS: Structure Identification and Modeling with a Self-Organizing Neuro-Fuzzy Inference System

AbstractThis paper presents a new adaptive learning algorithm to automatically design a neural fuzzy model. This constructive learning algorithm attempts to identify the structure of the model based on an architectural self-organization mechanism with a data-driven approach. The proposed training algorithm self-organizes the model with intuitive adding, merging and splitting operations. Sub-networks compete to learn specific training patterns and, to accomplish this task, the algorithm can either add new neurons, merge correlated ones or split existing ones with unsatisfactory performance. The proposed algorithm does not use a clustering method to partition the input-space like most of the state of the art algorithms. The proposed approach has been tested on well-known synthetic and real-world benchmark datasets. The experimental results show that our proposal is able to find the most suitable architecture with better results compared with those obtained with other methods from the literature.

ROBUST SIEVE BOOTSTRAP PREDICTION INTERVALS FOR CONTAMINATED TIME SERIES

Time series prediction is of primary importance in a variety of applications from several science fields, like engineering, finance, earth sciences, etc. Time series prediction can be divided in to two main tasks, point and interval estimation. Estimating prediction intervals, is in some cases more important than point estimation mainly because it indicates the likely uncertainty in the prediction process. Recently, the sieve bootstrap method has been successfully used in prediction of nonlinear time series. In this work, we study the performance of the prediction intervals based on the sieve bootstrap technique, which does not require the distributional assumption of normality as most techniques that are found in the literature. The construction of prediction intervals in the presence of different types of outliers is not robust from a distributional point of view, leading to an undesirable increase in the length of the prediction intervals. In the analysis of time series, it is common to have irregular observations that have different types of outliers. For this reason, we propose the construction of prediction intervals for returns based on the winsorized residual and bootstrap techniques for time series prediction. We propose a novel, simple and distribution free resampling technique for developing robust prediction intervals for returns and volatilities for GARCH models. The proposed procedure is illustrated by an application to known synthetic and real-time series.

Wind Speed Forecast under a Distributed Learning Approach

In this paper we apply a distributed learning approach to improve the perfomance of wind speed forecast. We use data obtained from 54 different weather stations in the U. S. and without sharing data between sites, we share model information between them, to improve the performance over local models trained with only local data. We show that sharing the information of the distributed models, improves the forecast we could obtain by only using locally trained models.

Sieve Bootstrap Prediction Intervals for Contaminated Non-linear Processes

Recently, the sieve bootstrap method has been successfully used in prediction of nonlinear time series. In this work we study the performance of the prediction intervals based on the sieve bootstrap technique, which does not require the distributional assumption of normality as most techniques that are found in the literature. The construction of prediction intervals in the presence of patchy outliers are not robust from a distributional point of view, leading to an undesirable increase in the length of the prediction intervals. n nIn the analysis of financial time series it is common to have irregular observations that have different types of outliers, isolated and in groups. For this reason we propose the construction of prediction intervals for returns based in the winsorized residual and bootstrap techniques for financial time series. We propose a novel, simple, efficient and distribution free resampling technique for developing robust prediction intervals for returns and volatilities for TGARCH models. The proposed procedure is illustrated by an application to known synthetic time series.

SIFAR: Self-Identification of Lags of an Autoregressive TSK-based Model

In this work, a Takagi-Sugeno-Kang (TSK) model is used for time series analysis and some important questions about the identification of this kind of models are addressed: the identification of the model structure and the set of the most influential regressors or lags. The main idea behind of the proposed method resembles to those techniques that prioritize lags evaluating the proximity of nearby samples in the input space in relation to the closeness of the corresponding target values. Clusters of samples are generated and the consistence of the mapping between the predicted variable and the set of candidate past values is evaluated. Afterwards, a TSK model is established and the redundancies in the rule base are avoided. Simulation experiments were conducted for 2 synthetic nonlinear autoregressive processes and for 4 benchmark time series. Results show a promising performance in terms of forecasting error and in terms of ability to find a proper set of lags of a given autoregressive process.