Search the repository

A+ | A- | | SLO | ENG

Abstract: Features are the main rule building blocks for rule learning algorithms. They can be simple tests for attribute values or complex logical terms representing available domain knowledge. In contrast to common practice in classification rule learning, we argue that separation of the feature construction and rule construction processes has theoretical and practical justification. Explicit usage of features enables a unifying framework of both propositional and relational rule learning and we present and analyze procedures for feature construction in both types of domains. It is demonstrated that the presented procedure for constructing a set of simple features has the property that the resulting set enables construction of complete and consistent rules whenever it is possible, and that the set does not include obviously irrelevant features. Additionally, the concept of feature relevancy is important for the effectiveness of rule learning. It this work, we illustrate the concept in the coverage space and prove that the relative relevancy has the quality-preserving property in respect to the resulting rules. Moreover, we show that the transformation from the attribute to the feature space enables a novel, theoretically justified way of handling unknown attribute values. The same approach enables that estimated imprecision of continuous attributes can be taken into account, resulting in construction of robust features in respect to this imprecision.
Found in: ključnih besedah
Summary of found: ... Machine learning, Feature construction, Rule learning, Unknown attribute...
Keywords: Machine learning, Feature construction, Rule learning, Unknown attribute values
Published: 14.07.2017; Views: 3708; Downloads: 0
Fulltext (365,76 KB)

Abstract: Cosmic rays with energies above 10^18 eV, usually referred to as ultra-high energy cosmic rays (UHECR), have been a mystery from the moment they have been discovered. Although we have now more information on their extragalactic origin, their direct sources still remain hidden due to deviations caused by galactic magnetic fields. Another mystery, apart from their production sites, is their nature. Their mass composition, still uncertain at these energies, would give us a better understanding on their production, acceleration, propagation and capacity to produce extensive air showers in the Earth's atmosphere. Mass composition studies of UHECR try to determine their nature from the difference in development of their extensive air showers. In this work, observational parameters from the hybrid detection system of the Pierre Auger Observatory are used in a multivariate analysis to obtain the mass composition of UHECR. The multivariate analysis (MVA) approach combines a number of mass composition sensitive variables and tries to improve the separation between different UHECR particle masses. Simulated distributions of different primary particles are fitted to measured observable distributions in order to determine individual elemental fractions of the composition. When including observables from the surface detector, we find a discrepancy in the estimated mass composition between a mixed simulation sample and the Pierre Auger data. Our analysis results from the Pierre Auger data are to a great degree independent on hadronic interaction models. Although they differ at higher primary masses, the different models are more consistent, when combining fractions of oxygen and iron. Compared to previously published results, the systematic uncertainty from hadronic interaction models is roughly four times smaller. Our analysis reports a predominantly heavy composition of UHECR, with more than a 50% fraction of oxygen and iron at low energies. The composition is then becoming heavier with increasing energy, with a fraction of oxygen and iron above 80% at the highest energies.
Found in: ključnih besedah
Summary of found: ...air showers, mass composition, Pierre Auger Observatory, machine learning, multivariate analysis...
Keywords: astroparticle physics, ultra-high energy cosmic rays, extensive air showers, mass composition, Pierre Auger Observatory, machine learning, multivariate analysis
Published: 03.04.2019; Views: 3922; Downloads: 166
Fulltext (17,53 MB)

Abstract: Epidemiological modeling is an important problem around the world. This research presents COVID-19 analysis to understand which model works better for different regions. A comparative analysis of three growth curve fitting models (Gompertz, Logistic, and Exponential), two mathematical models (SEIR and IDEA), two forecasting models (Holt’s exponential and ARIMA), and four machine/deep learning models (Neural Network, LSTM Networks, GANs, and Random Forest) using three evaluation criteria on ten prominent regions around the world from North America, South America, Europe, and Asia has been presented. The minimum and median values for RMSE were 1.8 and 5372.9; the values for the mean absolute percentage error were 0.005 and 6.63; and the values for AIC were 87.07 and 613.3, respectively, from a total of 125 experiments across 10 regions. The growth curve fitting models worked well where flattening of the cases has started. Based on region’s growth curve, a relevant model from the list can be used for predicting the number of infected cases for COVID-19. Some other models used in forecasting the number of cases have been added in the future work section, which can help researchers to forecast the number of cases in different regions of the world.
Found in: ključnih besedah
Summary of found: ...models (Holt’s exponential and ARIMA), and four machine/deep learning models (Neural Network, LSTM Networks, GANs,...
Keywords: epidemic modeling, machine learning, neural networks, pandemic forecasting, time-series forecasting
Published: 15.07.2021; Views: 1440; Downloads: 30
Fulltext (3,16 MB)
This document has many files! More...

Found in: ključnih besedah
Summary of found: ... machine learning, gamma rays...
Keywords: machine learning, gamma rays
Published: 17.02.2022; Views: 865; Downloads: 4
Fulltext (8,16 MB)
This document has many files! More...

Abstract: Objectives: The present work reports the study of 34 rhonchi (RB) and Bronchial Breath (BB) signals employing machine learning techniques, timefrequency, fractal, and non-linear time-series analyses. Methods: The timefrequency analyses and the complexity in the dynamics of airflow in BB and RB are studied using both Power Spectral Density (PSD) features and non-linear measures. For accurate prediction of these signals, PSD and nonlinear measures are fed as input attributes to various machine learning models. Findings: The spectral analyses reveal fewer, low-intensity frequency components along with its overtones in the intermittent and rapidly damping RB signal. The complexity in the dynamics of airflow in BB and RB is investigated through the fractal dimension, Hurst exponent, phase portrait, maximal Lyapunov exponent, and sample entropy values. The greater value of entropy for the RB signal provides an insight into the internal morphology of the airways containing mucous and other obstructions. The Principal Component Analysis (PCA) employs PSD features, and Linear Discriminant Analysis (LDA) along with Pattern Recognition Neural Network (PRNN) uses non-linear measures for predicting BB and RB. Signal classification based on phase portrait features evaluates the multidimensional aspects of signal intensities, whereas that based on PSD features considers mere signal intensities. The principal components in PCA cover about 86.5% of the overall variance of the data class, successfully distinguishing BB and RB signals. LDA and PRNN that use nonlinear time-series parameters identify and predict RB and BB signals with 100% accuracy, sensitivity, specificity, and precision. Novelty: The study divulges the potential of non-linear measures and PSD features in classifying these signals enabling its application to be extended for low-cost, non-invasive COVID-19 detection and real-time health monitoring.
Found in: ključnih besedah
Keywords: lung signal, fractal analysis, sample entropy, non­linear time­series, machine learning techniques
Published: 30.06.2022; Views: 559; Downloads: 0
Fulltext (1,50 MB)