Combinations Of Input Features Research Articles

Research on Digital Forensics Based on Uyghur Web Text Classification

This paper mainly discusses the use of mutual information (MI) and Support Vector Machines (SVMs) for Uyghur Web text classification and digital forensics process of web text categorization: automatic classification and identification, conversion and pretreatment of plain text based on encoding features of various existing Uyghur Web documents etc., introduces the pre-paratory work for Uyghur Web text encoding. Focusing on the non-Uyghur characters and stop words in the web texts filtering, we put forward a Multi-feature Space Normalized Mutual Information (M-FNMI) algorithm and replace MI between single feature and category with mutual information (MI) between input feature combination and category so as to extract more accurate feature words; finally, we classify features with support vector machine (SVM) algorithm. The experimental result shows that this scheme has a high precision of classification and can provide criterion for digital forensics with specific purpose.

Application of neural networks for evaluating energy performance certificates of residential buildings

The Energy Performance Building Directive 91 of 2002, mandates Member States of the European Union to enforce energy certification of buildings through local legislation. Among the Italian regions, Lombardy has issued predicted energy performance certificates for buildings since 2007 which accumulate to over one million entries. The current study is an attempt to validate a dataset of energy certificates by benefitting from the magnitude of registered buildings. Considering that manual evaluation of every entry is exhaustive and time consuming, artificial neural network is used as a fast and robust alternative for predicting heat demand indicators. Various combinations of input features are compared for selecting a reliable model. The number of inputs and hidden neurons are also optimized in order to achieve better accuracy. Results show that using 12 variables from an energy certificate is sufficient for estimating the related heat demand indicator. Regarding the stochastic initialization of neural networks, a set of 100 models are trained for obtaining a frequency distribution and confidence interval. Final results indicate that about 95% of entries fall within ±3 confidence intervals.

Integrated inference and learning of neural factors in structural support vector machines

Tackling pattern recognition problems in areas such as computer vision, bioinformatics, speech or text recognition is often done best by taking into account task-specific statistical relations between output variables. In structured prediction, this internal structure is used to predict multiple outputs simultaneously, leading to more accurate and coherent predictions. Structural support vector machines (SSVMs) are nonprobabilistic models that optimize a joint input–output function through margin-based learning. Because SSVMs generally disregard the interplay between unary and interaction factors during the training phase, final parameters are suboptimal. Moreover, its factors are often restricted to linear combinations of input features, limiting its generalization power. To improve prediction accuracy, this paper proposes: (i) joint inference and learning by integration of back-propagation and loss-augmented inference in SSVM subgradient descent; (ii) extending SSVM factors to neural networks that form highly nonlinear functions of input features. Image segmentation benchmark results demonstrate improvements over conventional SSVM training methods in terms of accuracy, highlighting the feasibility of end-to-end SSVM training with neural factors.

Joint Group Sparse PCA for Compressed Hyperspectral Imaging.

A sparse principal component analysis (PCA) seeks a sparse linear combination of input features (variables), so that the derived features still explain most of the variations in the data. A group sparse PCA introduces structural constraints on the features in seeking such a linear combination. Collectively, the derived principal components may still require measuring all the input features. We present a joint group sparse PCA (JGSPCA) algorithm, which forces the basic coefficients corresponding to a group of features to be jointly sparse. Joint sparsity ensures that the complete basis involves only a sparse set of input features, whereas the group sparsity ensures that the structural integrity of the features is maximally preserved. We evaluate the JGSPCA algorithm on the problems of compressed hyperspectral imaging and face recognition. Compressed sensing results show that the proposed method consistently outperforms sparse PCA and group sparse PCA in reconstructing the hyperspectral scenes of natural and man-made objects. The efficacy of the proposed compressed sensing method is further demonstrated in band selection for face recognition.

WE-AB-204-04: Feature Selection and Clustering Optimization for Pseudo-CT Generation in MR-Based Attenuation Correction and Radiation Therapy Planning

Purpose: Accurate and robust photon attenuation derived from MR is essential for PET/MR and MR-based radiation treatment planning applications. Although the fuzzy C-means (FCM) algorithm has been applied for pseudo-CT generation, the input feature combination and the number of clusters have not been optimized. This study aims to optimize both for clinically practical pseudo-CT generation. Methods: Nine volunteers were recruited. A 190-second, single-acquisition UTE-mDixon with 25% (angular) sampling and 3D radial readout was performed to acquire three primitive MR features at TEs of 0.1, 1.5, and 2.8 ms: the free-induction-decay (FID), the first and the second echo images. Three derived images, Dixon-fat and Dixon-water generated by two-point Dixon water/fat separation, and R2* (1/T2*) map, were also created. To identify informative inputs for generating a pseudo-CT image volume, all 63 combinations, choosing one to six of the feature images, were used as inputs to FCM for pseudo-CT generation. Further, the number of clusters was varied from four to seven to find the optimal approach. Mean prediction deviation (MPD), mean absolute prediction deviation (MAPD), and correlation coefficient (R) of different combinations were compared for feature selection. Results: Among the 63 feature combinations, the four that resulted in the best MAPD and R were further compared along with the set containing all six features. The results suggested that R2* and Dixon-water are the most informative features. Further, including FID also improved the performance of pseudo-CT generation. Consequently, the set containing FID, Dixon-water, and R2* resulted in the most accurate, robust pseudo-CT when the number of cluster equals to five (5C). The clusters were interpreted as air, fat, bone, brain, and fluid. The six-cluster Result additionally included bone marrow. Conclusion: The results suggested that FID, Dixon-water, R2* are the most important features. The findings can be used to facilitate pseudo-CT generation for unsupervised clustering. Please note that the project was completed with partial funding from the Ohio Department of Development grant TECH 11-063 and a sponsored research agreement with Philips Healthcare that is managed by Case Western Reserve University. As noted in the affiliations, some of the authors are Philips employees

Incorporation of texture information in a SVM method for classifying salt cedar in western China

Salt cedar is prominent plant species for wind-erosion resistance and saline–alkaline tolerance in arid desert and saline–alkali soil in China. Although some studies have been published for documenting salt cedar dynamics in its invasive habitat in the United States, the study of salt cedar in its native habitat, e.g. China, is relative scarce. Recently, the reduction of salt cedar is remarkable in western China that effective methods for monitoring salt cedar dynamics are essential for ecological environment preservation. As such, this study aims to develop a support vector machine (SVM) to better mapping salt cedar with high spatial resolution remote sensing image. Specifically, our objectives are twofold: (1) to examine the role that textures play in the SVM classification and further derive the optimal window size for extracting textures and (2) to determine optimal values for two vital parameters γ and C in the SVM method. To test the effectiveness of the developed SVM method, we compared the classification performance of SVM with three different combinations of input features, i.e. spectra only; spectra and textures; and spectra, textures, and Normalized Differential Vegetation Index (NDVI). In addition, for comparison purposes, a maximum likelihood classification method based on spectra, textures, and NDVI was adopted. Results indicate that (1) incorporation of textures and NDVI, especially textures, refines the SVM classification accuracy to 84.26%; (2) the optimal window size for texture calculation is 31 × 31 pixels; and (3) the optimal parameters of SVM model are C = 32 and γ = 8. Overall, the SVM method incorporating spectra, texture, and NDVI can accurately map salt cedar in our study site.

Mutual information based input feature selection for classification problems

The elimination process aims to reduce the size of the input feature set and at the same time to retain the class discriminatory information for classification problems. This paper investigates the approaches to solve classification problems of the feature selection and proposes a new feature selection algorithm using the mutual information (MI) concept in information theory for the classification problems. The proposed algorithm calculates the MI between the combinations of input features and the class instead of the MI between a single input feature and the class for both continuous-valued and discrete-valued features. Three experimental tests are conducted to evaluate the proposed algorithm. Comparison studies of the proposed algorithm with the previously published classification algorithms indicate that the proposed algorithm is robust, stable and efficient.

Spam detection using Random Boost

This paper proposes two alternative methods of random projections and compares their performance for robust and efficient spam detection when trained using a small number of examples. Robustness refers to learning and adaptation leading to a high level of performance despite data variability, while efficiency is concerned with (i) the complexity of the detection method employed; and (ii) the amount of training resources used for training and retraining. The first method, Random Project, employs a random projection matrix to produce linear combinations of input features, while the second method, Random Boost, employs random feature selection to enhance the performance of the Logit Boost algorithm. Random Boost is, in fact, a combination of Logit Boost and Random Forest. Experimental results, using TREC and CEAS as challenging spam benchmark sets, show that the Random Boost method significantly improves the performance of the spam filter compared to the Logit Boost algorithm (e.g., a 5% increase in AUC, which is the area under the Receiver Operating Characteristic curve), and yields similar classification accuracy compared to the Random Forest method but using only one fourth the runtime complexity of the Random Forest algorithm. Additionally, the Random Boost algorithm also reduces training time by two orders of magnitude compared to Logit Boost, which becomes important during retraining on the ever changing data streams, including adapting to adversarial tactics and “noise” injected by spammers.

Jordan recurrent neural network versus IHACRES in modelling daily streamflows

Summary A study of possible scenarios for modelling streamflow data from daily time series, using artificial neural networks (ANNs), is presented. Particular emphasis is devoted to the reconstruction of drought periods where water resource management and control are most critical. This paper considers two connectionist models: a feedforward multilayer perceptron (MLP) and a Jordan recurrent neural network (JNN), comparing network performance on real world data from two small catchments (192 and 69 km 2 in size) with irregular and torrential regimes. Several network configurations are tested to ensure a good combination of input features (rainfall and previous streamflow data) that capture the variability of the physical processes at work. Tapped delayed line (TDL) and memory effect techniques are introduced to recognize and reproduce temporal dependence. Results show a poor agreement when using TDL only, but a remarkable improvement can be obtained with JNN and its memory effect procedures, which are able to reproduce the system memory over a catchment in a more effective way. Furthermore, the IHACRES conceptual model, which relies on both rainfall and temperature input data, is introduced for comparative study. The results suggest that when good input data is unavailable, metric models perform better than conceptual ones and, in general, it is difficult to justify substantial conceptualization of complex processes.

Multisensor approach to automated classification of sea ice image data

A multisensor data fusion algorithm based on a multilayer neural network is presented for sea ice classification in the winter period. The algorithm uses European Remote Sensing (ERS), RADARSAT synthetic aperture radar (SAR), and low-resolution television camera images and image texture features. Based on a set of in situ observations made at the Kara Sea, a neural network is trained, and its structure is optimized using a pruning method. The performance of the algorithm with different combinations of input features (sensors) is assessed and compared with the performance of a linear discriminant analysis (LDA)-based algorithm. We show that for both algorithms a substantial improvement can be gained by fusion of the three different types of data (91.2% for the neural network) as compared with single-source ERS (66.0%) and RADARSAT (70.7%) SAR image classification. Incorporation of texture increases classification accuracy. This positive effect of texture becomes weaker with increasing number of sensors (from 8.4 to 6.4 percent points for the use of two and three sensors, respectively). In view of the short training time and smaller number of adjustable parameters, this result suggests that semiparametric classification methods can be considered as a good alternative to the neural networks and traditional parametric statistical classifiers applied for the sea ice classification.

PREDICTING APPLE BRUISING USING MACHINE LEARNING

Many models have been used to describe the influence of internal or external factors on apple bruising. Few of these have addressed the application of derived relationships to the evaluation of commercial operations. From an industry perspective, a model must enable fruit to be rejected on the basis of a commercially significant bruise and must also accurately quantify the effects of various combinations of input features (such as cultivar, maturity, size, and so on) on bruise prediction. Input features must in turn have characteristics which are measurable commercially; for example, the measure of force should be impact energy rather than energy absorbed. Further, as the commercial criteria for acceptable damage levels change, the model should be versatile enough to update bruise thresholds from existing data. Machine learning is a burgeoning technology with a vast range of potential applications particularly in agriculture where large amounts of data can be readily collected [1]. The main advantage of using a machine learning method in an application is that the models built for prediction can be viewed and understood by the owner of the data who is in a position to determine the usefulness of the model, an essential component in a commercial environment. Machine Learning software recently developed at Waikato University [2] offers potential as a prediction tool for the classification of bruising based on historical data. It gives the user the opportunity to select any number of measured input attributes and determine the influence of that combination on a range of bruise size categories. The user may require a high degree of accuracy in the classification and therefore prune the attributes or bruise classes accordingly, or alternatively seek to discover trends in the dataset (in which case a lower level of accuracy often clarifies implicit structures in the data). Models such as the theory of elasticity suggest that impact energy and radius of curvature will have a significant effect on the bruise surface area. Cell structure is also thought to contribute to variation in bruise size [3]. The experiment described in this paper uses the machine learning programs C4.5 [4] and M5' [5] to determine the influence of impact energy, radius of curvature and impact site location on bruise area.