Statistics for Spatial Data.

: This thesis applies neural network feature selection techniques to multivariate time series data to improve prediction of a target time series. Two approaches to feature selection are used. First, a subset enumeration method is used to determine which financial indicators are most useful for aiding in prediction of the S&P 500 futures daily price. The candidate indicators evaluated include RSI, Stochastics and several moving averages. Results indicate that the Stochastics and RSI indicators result in better prediction results than the moving averages. The second approach to feature selection is calculation of individual saliency metrics. A new decision boundary-based individual saliency metric, and a classifier independent saliency metric are developed and tested. Ruck's saliency metric, the decision boundary based saliency metric, and the classifier independent saliency metric are compared for a data set consisting of the RSI and Stochastics indicators as well as delayed closing price values. The decision based metric and the Ruck metric results are similar, but the classifier independent metric agrees with neither of the other metrics. The nine most salient features, determined by the decision boundary based metric, are used to train a neural network and the results are presented and compared to other published results. (AN)

/pdf/nonlinear-time-series-analysis-1j3a35n4y3.pdf

Nonlinear Time Series Analysis.

Generalized Linear Models (2nd ed.)

http://ieeexplore.ieee.org/iel5/5/31047/01444179.pdf

Fundamentals of acoustics

Hyperspectral sensors record the reflectance from the Earth's surface over the full range of solar wavelengths with high spectral resolution. The resulting high-dimensional data contain rich information for a wide range of applications. However, for a specific application, not all the measurements are important and useful. The original feature space may not be the most effective space for representing the data. Feature mining, which includes feature generation, feature selection (FS), and feature extraction (FE), is a critical task for hyperspectral data classification. Significant research effort has focused on this issue since hyperspectral data became available in the late 1980s. The feature mining techniques which have been developed include supervised and unsupervised, parametric and nonparametric, linear and nonlinear methods, which all seek to identify the informative subspace. This paper provides an overview of both conventional and advanced feature reduction methods, with details on a few techniques that are commonly used for analysis of hyperspectral data. A general form that represents several linear and nonlinear FE methods is also presented. Experiments using two widely available hyperspectral data sets are included to illustrate selected FS and FE methods.

Feature Mining for Hyperspectral Image Classification

Distributed detection in a large wireless sensor network

Feature selection is a key task in remote sensing data processing, particularly in case of classification from hyperspectral images. A logistic regression (LR) model may be used to predict the probabilities of the classes on the basis of the input features, after ranking them according to their relative importance. In this letter, the LR model is applied for both the feature selection and the classification of remotely sensed images, where more informative soft classifications are produced naturally. The results indicate that, with fewer restrictive assumptions, the LR model is able to reduce the features substantially without any significant decrease in the classification accuracy of both the soft and hard classifications

Logistic Regression for Feature Selection and Soft Classification of Remote Sensing Data

This paper introduces an online sleep apnea detection method based on heart rate complexity as measured by recurrence quantification analysis (RQA) statistics of heart rate variability (HRV) data. RQA statistics can capture nonlinear dynamics of a complex cardiorespiratory system during obstructive sleep apnea. In order to obtain a more robust measurement of the nonstationarity of the cardiorespiratory system, we use different fixed amount of neighbor thresholdings for recurrence plot calculation. We integrate a feature selection algorithm based on conditional mutual information to select the most informative RQA features for classification, and hence, to speed up the real-time classification process without degrading the performance of the system. Two types of binary classifiers, i.e., support vector machine and neural network, are used to differentiate apnea from normal sleep. A soft decision fusion rule is developed to combine the results of these classifiers in order to improve the classification performance of the whole system. Experimental results show that our proposed method achieves better classification results compared with the previous recurrence analysis-based approach. We also show that our method is flexible and a strong candidate for a real efficient sleep apnea detection system.

An Online Sleep Apnea Detection Method Based on Recurrence Quantification Analysis

Traditional biometric recognition systems often utilize physiological traits such as fingerprint, face, iris, etc. Recent years have seen a growing interest in electrocardiogram (ECG)-based biometric recognition techniques, especially in the field of clinical medicine. In existing ECG-based biometric recognition methods, feature extraction and classifier design are usually performed separately. In this paper, a multitask learning approach is proposed, in which feature extraction and classifier design are carried out simultaneously. Weights are assigned to the features within the kernel of each task. We decompose the matrix consisting of all the feature weights into sparse and low-rank components. The sparse component determines the features that are relevant to identify each individual, and the low-rank component determines the common feature subspace that is relevant to identify all the subjects. A fast optimization algorithm is developed, which requires only the first-order information. The performance of the proposed approach is demonstrated through experiments using the MIT-BIH Normal Sinus Rhythm database.

Joint Feature Extraction and Classifier Design for ECG-Based Biometric Recognition

Existing studies on the classical distributed detection problem typically assume idealized transmissions between local sensors and a fusion center. This is not guaranteed in the emerging wireless sensor networks with low-cost sensors and stringent power/delay constraints. By focusing on discrete transmission channels, we study the performance limits, in both asymptotic and non-asymptotic regimes, of a distributed detection system as a function of channel characteristics. For asymptotic analysis, we compute the error exponents of the underlying hypothesis testing problem; while for cases with a finite number of sensors, we determine channel conditions under which the distributed detection systems become useless - observing the channel outputs cannot help reduce the error probability at the fusion center. We demonstrate that as the number of sensors or the quantization levels at local sensors increase, the requirements on channel quality can be relaxed

/pdf/detection-performance-limits-for-distributed-sensor-networks-3jqhozlli2.pdf

Qi Cheng

Papers

Distributed detection in a large wireless sensor network

Logistic Regression for Feature Selection and Soft Classification of Remote Sensing Data

An Online Sleep Apnea Detection Method Based on Recurrence Quantification Analysis

Joint Feature Extraction and Classifier Design for ECG-Based Biometric Recognition

Detection Performance Limits for Distributed Sensor Networks in the Presence of Nonideal Channels