Selected Data Features Research Articles

Indoor human action recognition based on millimeter-wave radar micro-doppler signature

Considering that millimeter-wave radar lacks sufficient data to support Transfer Learning (TL) and Human Action Recognition (HAR), we propose a Heterogeneous Multi-source Transfer Learning (HMTL) method and a data selection algorithm based on Categorical Probability to obtain Hash Coding (CPHC). CPHC is utilized to select data from multi-source datasets in the same domain for downscaling and matching to ensure the similarity between the selected data features and the target task features to obtain better performance on the target task. The experimental results show that the CPHC can downscale data more efficiently than the traditional algorithm. HMTL can effectively improve the classification accuracy of the network to the state-of-the-art (SOAT) level. RPME addresses that the radar Micro-Doppler Signature (MDS) is flooded by noise due to the cluttered indoor environment, making the MDS more obvious and thus improving the classification accuracy.

A Kinematic Data-Driven Approach to Differentiate Involuntary Choreic Movements in Individuals With Neurological Conditions.

The ability to differentiate similar choreic involuntary movements could lay the groundwork for the development of a minimally-invasive screening tool for their etiology and provide in-depth understandings of pathophysiology. As a first step, we investigate kinematic differences between Huntington's disease (HD) chorea and Parkinson's disease (PD) choreic levodopa-induced dyskinesia (LID), which have distinct pathological causes yet share a great kinematic resemblance. Twenty subjects with HD and ten subjects with PD stood with both upper limbs in front of them for approximately 60 seconds. The three-dimensional velocity time-series of involuntary movements of both hands were segmented into one-dimensional sub-movements abutted by velocity zero-crossings. A combination of unsupervised and supervised machine learning algorithms was employed to automatically select data features extracted from sub-movements and distinguish the two types of involuntary choreic movements. The trained model was able to accurately classify chorea vs. LID with an Area Under the Receiver Operating Characteristic Curve of 99.5%. A set of important features contributing to the construction of the classification model were identified and investigated. The trained model may serve as a tool for the automatic identification of different types of involuntary choreic movements, enabling continuous monitoring and personalized treatment for patients in various clinical settings. The results provide insights into kinematic characteristics of HD chorea and PD LID, which is the first step towards an improved general understanding of involuntary choreic movements.

A Data-Driven Semi-Supervised Soft-Sensor Method: Application on an Industrial Cracking Furnace

The cracking furnace is the key equipment of the ethylene unit. Coking in furnace tubes results from the generation of coke during cracking, which will compromise the heat transfer efficiency and lead to shape change of tubes. In order to keep the cracking furnace operating economically and safely, the engineers need to decoke according to the surface temperature of the furnace tube. However, the surface temperature of the furnace tube is difficult to obtain in practice. Due to redundant instrumentation and the high level of process control in cracking furnaces, a large number of operation data have been collected, which makes it possible to predict the surface temperature of furnace tubes based on autocorrelation and cross correlation within and among variables. Traditional prediction methods rely on labeled data samples for training, ignoring the process information contained in a vast amount of unlabeled data. In this work, a data-driven semi-supervised soft-sensor method is proposed. Considering the nonlinear and dynamic relationship among variables, long short-term memory network (LSTM) autoencoder (AE), a deep neural network suitable for the feature extraction of long-term nonlinear series, is used for pre-training to extract process data features from unlabeled and labeled data. Then, principal component analysis (PCA) and mutual information (MI) are applied to remove feature correlation and select features related to target variables, respectively. Finally, the selected data features are utilized to establish a soft-sensor model based on artificial neural network (ANN). Data from an industrial cracking furnace of an ethylene unit is considered to validate the performance of the proposed method. The results show that the prediction error of furnace tube surface temperature is about 1% and successfully aid engineers in determining the optimal time for decoking.

A homogeneous ensemble based dynamic artificial neural network for solving the intrusion detection problem

Network security is a mechanism of protecting the usability and integrity of any given network and its transmitted data. Network security's effectiveness is crucial to the network environment to ensure it is free from any threat, especially in the critical infrastructure (CI). The supervisory control and data acquisition systems in the CI are getting more connected to the internet, putting them in serious security concerns. Any malicious attack against these systems could cause considerable human, economic, and material damages. Thus, it leads to the emergence of the intrusion detection system (IDS). Theoretically, a modern IDS must handle a large amount of data with high accuracy. Ensemble-based, hybrid-based methods and their distinguished applications are a promising way to solve these issues. The efficiency of the IDS is mainly dependent on the selected data features and the used classification method. The artificial neural network (ANN) has been applied in various fields, but it requires adjustment on few parameters to work effectively. This study proposes a homogeneous ensemble based on single-class dynamic ANN (HOE-DANN). Each dynamic ANN (DANN) is optimized by a filter-wrapper method using a modified discrete cuttlefish algorithm based on rough set theory, and a migration-strategy based cuttlefish algorithm. Both algorithms simultaneously optimize the features, ANN structure, weights, and biases for creating the DANN. However, the threshold value of the ensemble model was set using the hill-climbing algorithm. The experiments were applied to well-known benchmark datasets, namely the KDD99, UNSW-NB15, and gas pipeline data logs (GPDL). The results show that the HOE-DANN outperforms the single model based on the DANN. Additionally, a comparison with several state-of-the-art methods has shown that the proposed method offers superior performance in terms of the detection rate (DR), false alarm rate (FAR), and classification accuracy (ACC). The HOE-DANN model was able to achieve DR of 97.47%, FAR of 2.25%, and ACC of 97.52% using the KDD99 dataset, DR of 99.93%, FAR of 13.13%, and ACC of 94.08% using the UNSW-NB15 dataset, and DR of 98.08%, FAR of 2.69%, and ACC of 94.50% using the GPDL dataset.

MRMD2.0: A Python Tool for Machine Learning with Feature Ranking and Reduction

Aims: The study aims to find a way to reduce the dimensionality of the dataset. Background: Dimensionality reduction is the key issue of the machine learning process. It does not only improve the prediction performance but also could recommend the intrinsic features and help to explore the biological expression of the machine learning “black box”. Objective: A variety of feature selection algorithms are used to select data features to achieve dimensionality reduction. Methods: First, MRMD2.0 integrated 7 different popular feature ranking algorithms with PageRank strategy. Second, optimized dimensionality was detected with forward adding strategy. Result: We have achieved good results in our experiments. Conclusion: Several works have been tested with MRMD2.0. It showed well performance. Otherwise, it also can draw the performance curves according to the feature dimensionality. If users want to sacrifice accuracy for fewer features, they can select the dimensionality from the performance curves. Other: We developed friendly python tools together with the web server. The users could upload their csv, arff or libsvm format files. Then the webserver would help to rank features and find the optimized dimensionality.

Training deep neural density estimators to identify mechanistic models of neural dynamics.

Mechanistic modeling in neuroscience aims to explain observed phenomena in terms of underlying causes. However, determining which model parameters agree with complex and stochastic neural data presents a significant challenge. We address this challenge with a machine learning tool which uses deep neural density estimators-trained using model simulations-to carry out Bayesian inference and retrieve the full space of parameters compatible with raw data or selected data features. Our method is scalable in parameters and data features and can rapidly analyze new data after initial training. We demonstrate the power and flexibility of our approach on receptive fields, ion channels, and Hodgkin-Huxley models. We also characterize the space of circuit configurations giving rise to rhythmic activity in the crustacean stomatogastric ganglion, and use these results to derive hypotheses for underlying compensation mechanisms. Our approach will help close the gap between data-driven and theory-driven models of neural dynamics.

From affect, behavior, and cognition to personality: an integrated personal character model for individual-like intelligent artifacts

An individual-like intelligent artifact is a special kind of humanoid which resembles a human being in assimilating aspects of its real human counterpart’s cognition and neurological functions. Such an individual-like intelligent artifact could have a number of far-reaching applications, such as in creating a digital clone of an individual and bringing about forms of digital immortality. Although such intelligent artifacts have been created in various forms, such as physical robots or digital avatars, these creations are still far from modeling the inner cognitive and neurological mechanisms of an individual human. To imbue individual-like intelligent artifacts with the characteristics of individuals, we propose a Personal Character Model that consists of personality, the characteristics of affect, behavior, and cognition, and the relations between these characteristics. According to differential psychology and personality psychology, personality is the set of essential characteristics that make a person unique whereas characteristics in affect, behavior, and cognition explain a person’s stable and abstract personality in their diverse daily behavior. In addition, relations among these characteristics serve as a bridge from one characteristic to another. To illustrate the computing process of the personal character model, we first designed three experiments to collect physiological data and behavior data from twenty participants. Then we selected data features from the collected data using correlational analysis. Finally, we computed several representative characteristics from selected data features and represented the computed results.

Irradiance prediction intervals for PV stochastic generation in microgrid applications

The increasing interest in integrating volatile resources into microgrids implies the necessity of quantifying the uncertainty of photovoltaic (PV) production using dedicated probabilistic forecast techniques. The work presents a novel method to construct ultra-short-term and short-term prediction intervals (PIs) for solar global horizontal irradiance (GHI). The model applies the k-means algorithm to cluster observations of the clear-sky index according to the value of selected data features. At each timestep, the features are compared with the actual conditions to identify the representative cluster. The lower and upper bounds of the PI are calculated as the quantiles of the irradiance instances belonging to the selected cluster at a target confidence level. The validation is performed in 3 datasets of GHI measurements, each one of 85days. The model is able to deliver high performance PIs for forecast horizons ranging from sub-second to intra-hour ahead without the need of additional sensing systems such as all-sky cameras.

Computational inference of a genomic pluripotency signature in human and mouse stem cells.

Recent analyses of next-generation sequencing datasets have shown that cell-specific regulatory elements in stem cells are marked with distinguishable patterns of transcription factor (TF) binding and epigenetic marks. For example, we recently demonstrated that promoters of cell-specific genes are covered with expanded trimethylation of histone H3 at lysine 4 (H3K4me3) marks (i.e., broad H3K4me3 domains). Moreover, binding of specific TFs, such as OCT4, NANOG, and SOX2, have been shown to play a critical role in maintaining the pluripotency of stem cells. Despite these observations, a systematic exploration of genomic and epigenomic features of stem-cell-specific gene promoters has not been conducted. Advanced machine-learning models can capture distinguishable genomic and epigenomic characteristics of stem-cell-specific promoters by taking advantage of the wealth of publicly available datasets. Here, we propose a three-step framework to discover novel data characteristics of high-throughput next generation sequencing datasets that distinguish pluripotency genes in human and mouse embryonic stem cells (ESCs). Our framework involves: i) feature extraction to identify novel features of genomic datasets; ii) feature selection using a logistic regression model combined with the Least Absolute Shrinkage and Selection Operator (LASSO) method to find the most critical datasets and features; and iii) cross validation with features selected using LASSO method to assess the predictive power of selected data features in distinguishing pluripotency genes. We show that specific epigenetic marks, and specific features of these marks, are enriched at pluripotency gene promoters. Moreover, we also assess both the individual and combined effect of TF binding, epigenetic mark deposition, gene expression datasets for marking pluripotency genes. Our findings are consistent with the existence of a conserved, complex and integrative genomic signature in ESCs that can be exploited to flag important candidate pluripotency genes. They also validate our computational framework for fostering a deeper understanding of genomic datasets in stem cells, in the future, could be extended to study cell-type-specific genomic landscapes in other cell types.Reviewers: This article was reviewed by Zoltan Gaspari and Piotr Zielenkiewicz.Electronic supplementary materialThe online version of this article (doi:10.1186/s13062-016-0148-z) contains supplementary material, which is available to authorized users.

Dimensionality Reduction for Probabilistic Neural Network in Medical Data Classification Problems

Abstract This article presents the study regarding the problem of dimensionality reduction in training data sets used for classification tasks performed by the probabilistic neural network (PNN). Two methods for this purpose are proposed. The first solution is based on the feature selection approach where a single decision tree and a random forest algorithm are adopted to select data features. The second solution relies on applying the feature extraction procedure which utilizes the principal component analysis algorithm. Depending on the form of the smoothing parameter, different types of PNN models are explored. The prediction ability of PNNs trained on original and reduced data sets is determined with the use of a 10-fold cross validation procedure.

Human Activity Recognition in WSN: A Comparative Study

Human activity recognition is an emerging field of ubiquitous and pervasive computing. Although recent smartphones have powerful resources, the execution of machine learning algorithms on a large amount of data is still a burden on smartphones. Three major factors including; classification algorithm, data feature, and smartphone position influence the recognition accuracy and time. In this paper, we present a comparative study of six classification algorithms, six data features, and four different positions that are most commonly used in the recognition process using smartphone accelerometer. This analysis can be used to select any specific classification algorithm, data feature, and smartphone position for human activity recognition in terms of accuracy and response time. The methodology we used is composed of two major components; a data collector, and a classifier. A set of eleven activities of daily living, four different positions for data collection and ten volunteers contributed to make it a worth-full comparative study. Results show that K-Nearest Neighbor and J48 algorithms performed well both in terms of time and accuracy irrespective of data features whereas the performance of other algorithms is dependent on the selected data features. Similarly, mean and mode features gave good results in terms of accuracy irrespective of the classification algorithm. A short version of the paper has already been presented at ICIS 2014.

A Method For Knowledge Representation And Discovery Based On Composing And Manipulating Logical Equations

When information about objects and processes is formalised in knowledge bases, there is often the need to deal with features that have discrete values, for example gender (male/female) or marital status (married/single/divorced). When the mutual influence of discrete features is investigated, it is important to fmd out how closely the variables representing these features are connected and how strongly they influence each other. It is also important to obtain analytical dependencies in the data. This paper focuses on the logical correlation in knowledge bases, and suggests a method for knowledge representation and discovery of dependencies between selected data features. A knowledge base containing links between information features is represented in the form of logical equations with finite predicates. Special classes of equations describing complicated hierarchical data structures are considered. New knowledge about logical links in the data are obtained by eliminating variables from these equations through procedures involving the use of both quantifiers, and the role of each quantifier is discussed. The results of these procedures are dependencies between the features that are easier to interpret than the dependencies represented by the original equations. The elimination procedure does not lead to an increase in the size of the original knowledge base, and there are possibilities for parallel calculations. The knowledge base can be written in a text file with the help of a subset of the extensible mark-up language (XML), which makes it easy to transfer and manipulate the data. Some examples are discussed for the purpose of illustration.