Feature Generation Techniques Research Articles

Classification of Driver Distraction: A Comprehensive Analysis of Feature Generation, Machine Learning, and Input Measures.

The objective of this study was to analyze a set of driver performance and physiological data using advanced machine learning approaches, including feature generation, to determine the best-performing algorithms for detecting driver distraction and predicting the source of distraction. Distracted driving is a causal factor in many vehicle crashes, often resulting in injuries and deaths. As mobile devices and in-vehicle information systems become more prevalent, the ability to detect and mitigate driver distraction becomes more important. This study trained 21 algorithms to identify when drivers were distracted by secondary cognitive and texting tasks. The algorithms included physiological and driving behavioral input processed with a comprehensive feature generation package, Time Series Feature Extraction based on Scalable Hypothesis tests. Results showed that a Random Forest algorithm, trained using only driving behavior measures and excluding driver physiological data, was the highest-performing algorithm for accurately classifying driver distraction. The most important input measures identified were lane offset, speed, and steering, whereas the most important feature types were standard deviation, quantiles, and nonlinear transforms. This work suggests that distraction detection algorithms may be improved by considering ensemble machine learning algorithms that are trained with driving behavior measures and nonstandard features. In addition, the study presents several new indicators of distraction derived from speed and steering measures. Future development of distraction mitigation systems should focus on driver behavior-based algorithms that use complex feature generation techniques.

TV-Program Retrieval and Classification: A Comparison of Approaches based on Machine Learning

Electronic Program Guides (EPGs) are systems that allow users of media applications, such as web TVs, to navigate scheduling information about current and upcoming programming. Personalized EPGs help users to overcome information overload in this domain, by exploiting recommender systems that automatically compile lists of novel and diverse video assets, based on implicitly or explicitly defined user preferences. In this paper we introduce the concept of personal channel, on which Personalized EPGs are grounded, that provides users with potentially interesting programs and videos, by exploiting program genres (documentary, sports, …) and short textual descriptions of programs to find and categorize them. We investigate the problem of adopting appropriate algorithms for TV-program classification and retrieval, in the context of building personal channels, which is harder than a classical retrieval or classification task because of the short text available. The approach proposed to overcome this problem is the adoption of a new feature generation technique that enriches the textual program descriptions with additional features extracted from Wikipedia. Results of the experiments show that our approach actually improves the retrieval performance, while a limited positive effect is observed on classification accuracy.

A method for determining the number of features in the kernel space required for preserving classifiability

Feature generation techniques that sort the generated features in terms of their importance, such as principal component analysis (PCA), reduce the problem of feature subset selection to only determining the number of features to be retained. For databases with linearly inseparable classes, kernel PCA can be used as the feature generation method instead of the linear PCA. However, determining the number of features in the kernel space that needs to be retained for preserving the classifiability is a difficult problem, since the data vectors are not available in an explicit form in that space. In this paper, we propose a criterion and an algorithm for determining the number of required features in kernel PCA using only the elements of the kernel matrix. In order to show the effectiveness of the proposed criterion, the new algorithm is applied to the USPS handwritten digit, Yale Face and Caltech 101 databases. The proposed algorithm is also investigated for its robustness to noise that corrupts the data samples.

Automatic feature generation for machine learning--based optimising compilation

Recent work has shown that machine learning can automate and in some cases outperform handcrafted compiler optimisations. Central to such an approach is that machine learning techniques typically rely upon summaries or features of the program. The quality of these features is critical to the accuracy of the resulting machine learned algorithm; no machine learning method will work well with poorly chosen features. However, due to the size and complexity of programs, theoretically there are an infinite number of potential features to choose from. The compiler writer now has to expend effort in choosing the best features from this space. This article develops a novel mechanism to automatically find those features that most improve the quality of the machine learned heuristic. The feature space is described by a grammar and is then searched with genetic programming and predictive modelling. We apply this technique to loop unrolling in GCC 4.3.1 and evaluate our approach on a Pentium 6. On a benchmark suite of 57 programs, GCCs hard-coded heuristic achieves only 3% of the maximum performance available, whereas a state-of-the-art machine learning approach with hand-coded features obtains 59%. Our feature generation technique is able to achieve 76% of the maximum available speedup, outperforming existing approaches.

Faults Classification Scheme for Three Phase Induction Motor

In every kind of industrial application, the operation of fault detection and diagnosis for induction motors is of paramount importance. Fault diagnosis and detection led to minimize the downtime and improves its reliability and availability of the systems. In this article, a fault classification algorithm based on a robust linear discrimination scheme, for the case of a squirrel–cage three phase induction motor, will be presented. The suggested scheme is based on a novel feature extraction mechanism from the measured magnitude and phase of current park's vector pattern. The proposed classification algorithm is applied to detect of two kinds of induction machine faults, which area) broken rotor bar, and b) short circuit in stator winding. The novel feature generation technique is able to transform the problem of fault detection and diagnosis into a simpler space, where direct robust linear discrimination can be applied for solving the classification problem. And thus a clear classification of the healthy and the faulty cases can be robustly performed, by having the optimal hyper plane. This method can separate the feature current classes in a low dimensional subspace. Robust linear discrimination has been one of the most widely used fault detection methods in real-life applications, as this methodology seeks for directions that are efficient for discrimination and at the same time applies a straight-forward implementation. The efficacy of the proposed scheme will be evaluated based on multiple simulation results in different fault types.

Noise robust ASR in reverberated multisource environments applying convolutive NMF and Long Short-Term Memory

This article proposes and evaluates various methods to integrate the concept of bidirectional Long Short-Term Memory (BLSTM) temporal context modeling into a system for automatic speech recognition (ASR) in noisy and reverberated environments. Building on recent advances in Long Short-Term Memory architectures for ASR, we design a novel front-end for context-sensitive Tandem feature extraction and show how the Connectionist Temporal Classification approach can be used as a BLSTM-based back-end, alternatively to Hidden Markov Models (HMM). We combine context-sensitive BLSTM-based feature generation and speech decoding techniques with source separation by convolutive non-negative matrix factorization. Applying our speaker adapted multi-stream HMM framework that processes MFCC features from NMF-enhanced speech as well as word predictions obtained via BLSTM networks and non-negative sparse classification (NSC), we obtain an average accuracy of 91.86% on the PASCAL CHiME Challenge task at signal-to-noise ratios ranging from −6 to 9dB. To our knowledge, this is the best result ever reported for the CHiME Challenge task.

Automatic methods for characterization of sexual dimorphism of adult femora: distal femur

Quantifying sex differences in femoral size and shape has extensive applications in forensics and prosthesis design. By applying strong statistical techniques such as principal component analysis (PCA), certain three-dimensional (3D) morphological variations of adult femora can be quantified over various femoral sizes. Coupling this statistical approach with a novel feature generation and extraction technique, localization of statistically significant (p < 0.05) features are automatically defined and measured. Also, predefined anatomical landmarks and surgical axes have been calculated automatically. In all methods, femoral scale is controlled as a possible parameter of shape. By extensively comparing measurements across 92 male and 74 female femora, the dimorphic characteristics of the distal femur are shown. These differences have not been accounted for in many prosthetic systems and consequently these systems have limited sizing accuracy.

Structural feature indexing for retrieval of partially visible shapes

Efficient and robust information retrieval from large image databases is an essential functionality for the reuse, manipulation, and editing of multimedia documents. Structural feature indexing is a potential approach to efficient shape retrieval from large image databases, but the indexing is sensitive to noise, scales of observation, and local shape deformations. It has now been confirmed that efficiency of classification and robustness against noise and local shape transformations can be improved by the feature indexing approach incorporating shape feature generation techniques (Nishida, Comput. Vision Image Understanding 73 (1) (1999) 121–136). In this paper, based on this approach, an efficient, robust method is presented for retrieval of model shapes that have parts similar to the query shape presented to the image database. The effectiveness is confirmed by experimental trials with a large database of boundary contours obtained from real images, and is validated by systematically designed experiments with a large number of synthetic data.

Structural Shape Indexing with Feature Generation Models

Structural indexing is a potential approach to efficient classification and retrieval of image patterns with respect to a very large number of models. This technique is based on the idea of distributing features associated with model identifiers over a large data structure prepared for a model set, along with classification by voting for models with reference to the extracted features. Essential problems caused by mapping image features to discrete indices are that indexing is sensitive to noise, scales of observation, and local shape deformations, and thata prioriknowledge and feature distributions of corrupted instances are not available for each class when a large number of training data are not presented. To cope with these problems, shape feature generation techniques are incorporated into structural indexing. An analysis of feature transformations is carried out for some particular types of shape deformations, leading to feature generation rules composed of a small number of distinct cases. The rules are exploited to generate features that can be extracted from deformed patterns caused by noise and local shape deformations. In both processes of model database organization and classification, the generated features by the transformation rules are used for structural indexing and voting, as well as the features actually extracted from contours. The effectiveness of the proposed method is demonstrated by experimental trials with a large number of sample data. Furthermore, its application to shape retrieval from image databases is mentioned. The shape feature generation significantly improves the classification accuracy and efficiency.