Set Of Possible Features Research Articles

Artificial neural networks modeling of wall pressure spectra beneath turbulent boundary layers

We analyze and compare various empirical models of wall pressure spectra beneath turbulent boundary layers and propose an alternative machine learning approach using Artificial Neural Networks (ANNs). The analysis and the training of the ANN are performed on data from experiments and high-fidelity simulations by various authors, covering a wide range of flow conditions. We present a methodology to extract all the turbulent boundary layer parameters required by these models, also considering flows experiencing strong adverse pressure gradients. Moreover, the database is explored to unveil important dependencies within the boundary layer parameters and to propose a possible set of features from which the ANN should predict the wall pressure spectra. The results show that the ANN outperforms traditional models in adverse pressure gradients, and its predictive capabilities generalize better over the range of investigated conditions. The analysis is completed with a deep ensemble approach for quantifying the uncertainties in the model prediction and integrated gradient analysis of the model sensitivity to its inputs. Uncertainties and sensitivities allow for identifying the regions where new training data would be most beneficial to the model's accuracy, thus opening the path toward a self-calibrating modeling approach.

On Data-Dependent Random Features for Improved Generalization in Supervised Learning

The randomized-feature approach has been successfully employed in large-scale kernel approximation and supervised learning. The distribution from which the random features are drawn impacts the number of features required to efficiently perform a learning task. Recently, it has been shown that employing data-dependent randomization improves the performance in terms of the required number of random features. In this paper, we are concerned with the randomized-feature approach in supervised learning for good generalizability. We propose the Energy-based Exploration of Random Features (EERF) algorithm based on a data-dependent score function that explores the set of possible features and exploits the promising regions. We prove that the proposed score function with high probability recovers the spectrum of the best fit within the model class. Our empirical results on several benchmark datasets further verify that our method requires smaller number of random features to achieve a certain generalization error compared to the state-of-the-art while introducing negligible pre-processing overhead. EERF can be implemented in a few lines of code and requires no additional tuning parameters.

Sparse model selection via integral terms.

Model selection and parameter estimation are important for the effective integration of experimental data, scientific theory, and precise simulations. In this work, we develop a learning approach for the selection and identification of a dynamical system directly from noisy data. The learning is performed by extracting a small subset of important features from an overdetermined set of possible features using a nonconvex sparse regression model. The sparse regression model is constructed to fit the noisy data to the trajectory of the dynamical system while using the smallest number of active terms. Computational experiments detail the model's stability, robustness to noise, and recovery accuracy. Examples include nonlinear equations, population dynamics, chaotic systems, and fast-slow systems.

ASSESSMENT OF THE GEOMETRIC QUALITY OF SENTINEL-2 DATA

High resolution (10 m and 20 m) optical imagery satellite Sentinel-2 brings a new perspective to Earth observation. Its frequent revisit time enables monitoring the Earth surface with high reliability. Since Sentinel-2 data is provided free of charge by the European Space Agency, its mass use for variety of purposes is expected. Quality evaluation of Sentinel-2 data is thus necessary. Quality analysis in this experiment is based on comparison of Sentinel-2 imagery with reference data (orthophoto). From the possible set of features to compare (point features, texture lines, objects, etc.) line segments were chosen because visual analysis suggested that scale differences matter least for these features. The experiment was thus designed to compare long line segments (e.g. airstrips, roads, etc.) in both datasets as the most representative entities. Edge detection was applied to both images and corresponding edges were manually selected. The statistical parameter which describes the geometrical relation between different images (and between datasets in general) covering the same area is calculated as the distance between corresponding curves in two datasets. The experiment was conducted for two different test sites, Austria and Serbia. From 21 lines with a total length of ca. 120 km the average offset of 6.031 m (0.60 pixel of Sentinel-2) was obtained for Austria, whereas for Serbia the average offset of 12.720 m (1.27 pixel of Sentinel-2) was obtained out of 10 lines with a total length of ca. 38 km.

ASSESSMENT OF THE GEOMETRIC QUALITY OF SENTINEL-2 DATA

Abstract. High resolution (10 m and 20 m) optical imagery satellite Sentinel-2 brings a new perspective to Earth observation. Its frequent revisit time enables monitoring the Earth surface with high reliability. Since Sentinel-2 data is provided free of charge by the European Space Agency, its mass use for variety of purposes is expected. Quality evaluation of Sentinel-2 data is thus necessary. Quality analysis in this experiment is based on comparison of Sentinel-2 imagery with reference data (orthophoto). From the possible set of features to compare (point features, texture lines, objects, etc.) line segments were chosen because visual analysis suggested that scale differences matter least for these features. The experiment was thus designed to compare long line segments (e.g. airstrips, roads, etc.) in both datasets as the most representative entities. Edge detection was applied to both images and corresponding edges were manually selected. The statistical parameter which describes the geometrical relation between different images (and between datasets in general) covering the same area is calculated as the distance between corresponding curves in two datasets. The experiment was conducted for two different test sites, Austria and Serbia. From 21 lines with a total length of ca. 120 km the average offset of 6.031 m (0.60 pixel of Sentinel-2) was obtained for Austria, whereas for Serbia the average offset of 12.720 m (1.27 pixel of Sentinel-2) was obtained out of 10 lines with a total length of ca. 38 km.

Probability Density Function Estimation Using the EEF With Application to Subset/Feature Selection

We describe a method for estimating a probability density function when some of the sufficient statistics are known. The general form of the PDF is within the exponential family and is known as the exponentially embedded family. Using the proposed estimator new approaches to choosing features from a set of possible features become available, with the error metric being the Kullback-Liebler distance. Applications to subset selection in the context of multipath estimation as well as linear regression for machine learning are used to illustrate the practical utility of the method.

Features for stochastic approximation based foreground detection

Foreground detection algorithms have sometimes relied on rather ad hoc procedures, even when probabilistic mixture models are defined. Moreover, the fact that the input features have different variances and that they are not independent from each other is often neglected, which hampers performance. Here we aim to obtain a background model which is not tied to any particular choice of features, and that accounts for the variability and the dependences among features. It is based on the stochastic approximation framework. A possible set of features is presented, and their suitability for this problem is assessed. Finally, the proposed procedure is compared with several state-of-the-art alternatives, with satisfactory results.

Network Traffic Features for Anomaly Detection in Specific Industrial Control System Network

The deterministic and restricted nature of industrial control system networks sets them apart from more open networks, such as local area networks in office environments. This improves the usability of network security, monitoring approaches that would be less feasible in more open environments. One of such approaches is machine learning based anomaly detection. Without proper customization for the special requirements of the industrial control system network environment, many existing anomaly or misuse detection systems will perform sub-optimally. A machine learning based approach could reduce the amount of manual customization required for different industrial control system networks. In this paper we analyze a possible set of features to be used in a machine learning based anomaly detection system in the real world industrial control system network environment under investigation. The network under investigation is represented by architectural drawing and results derived from network trace analysis. The network trace is captured from a live running industrial process control network and includes both control data and the data flowing between the control network and the office network. We limit the investigation to the IP traffic in the traces.

Comparative analysis of mapping burned areas from landsat TM images

Remote sensing is a major source of mapping the burned area caused by forest fire. The focus in this application is to map a single class of interest, i.e. burned area. In this study, three different data combinations were classified using different classifiers and quantitatively compared. The adopted classifiers are Support Vector Data Description (SVDD), a one-class classifier, Binary classifier Support Vector Machines (SVMs) and traditional Maximum Likelihood classifier (ML). At first, the Principal Component Analysis (PCA) was applied to extract the best possible features form the original multispectral image (OMI) and calculated spectral indices (SI). Then the resulting subset of features was applied to the classifiers. The comparative study has undertaken to find firstly, the best possible set of features (data combination) and secondly, an effective classifier to map the burned areas. The best possible set of features was attained by data combination- II (i.e., OMI information). Furthermore, the results of the SVM showed the high classification accuracies than ML. Experimental results demonstrate that even though the SVDD for mapping the burned areas doesn't showed the higher classification accuracy than SVM, but it shows the suitability for the cases with few or poorly represented labelled samples available. The parameters should be further optimized through the use of intelligent training for improving the accuracy of SVDD.

Optimal feature selection for classifying a large set of chemicals using metal oxide sensors

Using linear support vector machines, we investigated the feature selection problem for the application of all-against-all classification of a set of 20 chemicals using two types of sensors, classical doped tin oxide and zeolite-coated chromium titanium oxide sensors. We defined a simple set of possible features, namely the identity of the sensors and the sampling times and tested all possible combinations of such features in a wrapper approach. We confirmed that performance is improved, relative to previous results using this data set, by exhaustive comparison of these feature sets. Using the maximal number of different sensors and all available data points for each sensor does not necessarily yield the best results, even for the large number of classes in this problem. We contrast this analysis, using exhaustive screening of simple feature sets, with a number of more complex feature choices and find that subsampled sets of simple features can perform better. Analysis of potential predictors of classification performance revealed some relevance of clustering properties of the data and of correlations among sensor responses but failed to identify a single measure to predict classification success, reinforcing the relevance of the wrapper approach used. Comparison of the two sensor technologies showed that, in isolation, the doped tin oxide sensors performed better than the zeolite-coated chromium titanium oxide sensors but that mixed arrays, combining both technologies, performed best.

PARTICLE SWARM OPTIMIZATION-BASED FEATURE SELECTION AND PARAMETER OPTIMIZATION FOR POWER SYSTEM DISTURBANCES CLASSIFICATION

In many data mining applications that address classification problems, feature and model selection are considered as key tasks. The appropriate input features of the classifier are selected from a given set of possible features, and the structure parameters of the classifier are adapted with respect to these features and a given dataset. This paper describes the particle swarm optimization algorithm (PSO) that performs feature and model selection simultaneously for the probabilistic neural network (PNN) classifier for power system disturbances. The probabilistic neural network is one of the successful classifiers used to solve many classification problems. However, the computational effort and storage requirement of the PNN method will prohibitively increase as the number of patterns used in the training set increases. An important issue that has not been given enough attention is the selection of a “spread parameter,” also called a “smoothing parameter,” in the PNN classifier. PSO is a powerful meta-heuristic technique in the artificial intelligence field; therefore, this study proposes a PSO-based approach, called PSO-PNN, to specify the beneficial features and the value of spread parameter to enhance the performance of PNN. The experimental results indicate that the proposed PSO-based approach significantly improves the classification accuracy with the discriminating input features for PNN.

A Hybrid Multilayer Perceptron Neural Network for Direct Marketing

Data Mining is the use of algorithms to extract the information and patterns derived by the knowledge discovery in database process. It is often referred to as supervised learning because the classes are determined before examining the data. In many data mining applications that address classification problems, feature and model selection are considered as key tasks. That is, appropriate input features of the classifier must be selected from a given set of possible features and structure parameters of the classifier must be adapted with respect to these features and a given data set. This paper describes feature selection and model selection simultaneously for Multilayer Perceptron (MLP) classifiers. In order to reduce the optimization effort, various techniques are integrated that accelerate and improve the classifier significantly. The feasibility and the benefits of the proposed approach are demonstrated by means of data mining problem: Direct Marketing in Customer Relationship Management. It is shown that, compared to earlier MLP technique, the run time is reduced with respect to learning data and with validation data for the proposed Multilayer Perceptron (MLP) classifiers. Similarly, the error rate is relatively low with respect to learning data and with validation data in direct marketing dataset. The algorithm is independent of specific applications so that many ideas and solutions can be transferred to other classifier paradigms.

Large-scale risk prediction applied to Genetic Analysis Workshop 17 mini-exome sequence data

We consider the application of Efron’s empirical Bayes classification method to risk prediction in a genome-wide association study using the Genetic Analysis Workshop 17 (GAW17) data. A major advantage of using this method is that the effect size distribution for the set of possible features is empirically estimated and that all subsequent parameter estimation and risk prediction is guided by this distribution. Here, we generalize Efron’s method to allow for some of the peculiarities of the GAW17 data. In particular, we introduce two ways to extend Efron’s model: a weighted empirical Bayes model and a joint covariance model that allows the model to properly incorporate the annotation information of single-nucleotide polymorphisms (SNPs). In the course of our analysis, we examine several aspects of the possible simulation model, including the identity of the most important genes, the differing effects of synonymous and nonsynonymous SNPs, and the relative roles of covariates and genes in conferring disease risk. Finally, we compare the three methods to each other and to other classifiers (random forest and neural network).

Form Defects Tolerancing by Natural Modes Analysis

The form defects quality needs methods to express allowable deviations. We propose a new language for form defects expression. This one is based on natural mode shapes of a discretized feature. The finite element method is used in order to compute those modes. Then a basis of defects is built with the natural modes. A defect is projected in this basis and thus the coordinates (modal coefficient) represent it. Hence, tolerancing is possible, by limiting those coordinates. The methods proposed in the literature can be applied on elementary geometries or there is a need to express the set of possible features (explicit geometry). Our method is versatile because it is based on the discretization of the feature (implicit geometry). The modal tolerancing method proposes two ways to express specifications of form defects: (1) The spectral tolerancing shows the modal coordinates and their limits in a bar chart graph by drawing the limits. In this method, we can see the decomposition of the measured feature and express tolerancing on each coordinate. (2) When a specification needs to link coordinates, we propose the modal domain method. An inclusion test of the feature coordinates gives the result of the metrology. Those methods are presented in an example.

Modeling timbre distance with temporal statistics from polyphonic music

Timbre distance and similarity are expressions of the phenomenon that some music appears similar while other songs sound very different to us. The notion of genre is often used to categorize music, but songs from a single genre do not necessarily sound similar and vice versa. In this work, we analyze and compare a large amount of different audio features and psychoacoustic variants thereof for the purpose of modeling timbre distance. The sound of polyphonic music is commonly described by extracting audio features on short time windows during which the sound is assumed to be stationary. The resulting down sampled time series are aggregated to form a high-level feature vector describing the music. We generated high-level features by systematically applying static and temporal statistics for aggregation. The temporal structure of features in particular has previously been largely neglected. A novel supervised feature selection method is applied to the huge set of possible features. The distances of the selected feature correspond to timbre differences in music. The features show few redundancies and have high potential for explaining possible clusters. They outperform seven other previously proposed feature sets on several datasets with respect to the separation of the known groups of timbrally different music.

Evolutionary Optimization of Radial Basis Function Classifiers for Data Mining Applications

In many data mining applications that address classification problems, feature and model selection are considered as key tasks. That is, appropriate input features of the classifier must be selected from a given (and often large) set of possible features and structure parameters of the classifier must be adapted with respect to these features and a given data set. This paper describes an evolutionary algorithm (EA) that performs feature and model selection simultaneously for radial basis function (RBF) classifiers. In order to reduce the optimization effort, various techniques are integrated that accelerate and improve the EA significantly: hybrid training of RBF networks, lazy evaluation, consideration of soft constraints by means of penalty terms, and temperature-based adaptive control of the EA. The feasibility and the benefits of the approach are demonstrated by means of four data mining problems: intrusion detection in computer networks, biometric signature verification, customer acquisition with direct marketing methods, and optimization of chemical production processes. It is shown that, compared to earlier EA-based RBF optimization techniques, the runtime is reduced by up to 99% while error rates are lowered by up to 86%, depending on the application. The algorithm is independent of specific applications so that many ideas and solutions can be transferred to other classifier paradigms.

Genetic algorithms for feature selection in machine condition monitoring with vibration signals

Artificial neural networks (ANNs) can be used successfully to detect faults in rotating machinery, using statistical estimates of the vibration signal as input features. In any given scenario, there are many different possible features that may be used as inputs for the ANN. One of the main problems facing the use of ANNs is the selection of the best inputs to the ANN, allowing the creation of compact, highly accurate networks that require comparatively little preprocessing. The paper examines the use of a genetic algorithm (GA) to select the most significant input features from a large set of possible features in machine condition monitoring contexts. Using a GA, a subset of six input features is selected from a set of 66, giving a classification accuracy of 99.8%, compared with an accuracy of 87.2% using an ANN without feature selection and all 66 inputs. From a larger set of 156 different features, the GA is able to select a set of six features to give 100% recognition accuracy.