Giovanni De Poli · 2014
Models for the representation the information from sound are necessary for the description of the in-formation, from the perceptive and operative point of view. Beyond the models, analysis methods are needed to discover the parameters which allow sound description, possibly lossless from the physical and perceptual properties description. When aiming to the extraction of information for sound, we need to discard every feature which is non relevant. This process of feature extraction consists of var-ious steps, starting from pre-processing the sound, then windowing, extraction, and post-processing procedures. An audio signal classification system can be generally represented as represented in Fig-ure 4.1. Pre-processing: The pre-processing consists of noise reduction, equalization, low-pass filtering. In speech processing (voice has a low-pass behavior) a pre-emphasis is applied by high-pass fil-tering the signal to smooth the spectrum, to achieve an uniform energy distribution spectrum. Windowing: Second step is to create frames from a continuous waveform signal. Frames are partially overlapping, as discuss in section 4.1.1.1 Extraction: Third step consists in obtaining a vector of features for each frame. Post-processing: In the post-processing phase, the most relevant cues of the features vector are se-lected. For instance, earlier mel-cepstral coefficients can be lightly weighted when having low frequency noise. 4.1.1 Time domain: Short-time analysis Time domain methods divide according to the time extent of the description provided by the feature. Short and long time analysis techniques apply to different parts of the object (i.e. a sounding note) to describe the attack of the sound (short extent) or the loudness of a note (long extent).
Cited by 0
Tahmina Afrose Keya (20141592), Siventhiran S. Balakrishnan (20141595), Maheswaran Solayappan (20141598) et al. · Figshare · 2024
<div><p>Malaysia, particularly Pahang, experiences devastating floods annually, causing significant damage. The objective of the research was to create a flood susceptibility map for the designated area by employing an Ensemble Machine Learning (EML) algorithm based on geographic information system (GIS). By analyzing nine key factors from a geospatial database, flood susceptibility map was created with the ArcGIS software (ESRI ArcGIS Pro v3.0.1 x64). The Random Forest (RF) model was employed in this study to categorize the study area into distinct flood susceptibility classes. The Feature selection (FS) method was used to ranking the flood influencing factors. To validate the flood susceptibility models, standard statistical measures and the Area Under the Curve (AUC) were employed. The FS ranking demonstrated that the primary attributes to flooding in the study region are rainfall and elevation, with slope, geology, curvature, flow accumulation, flow direction, distance from the river, and land use/land cover (LULC) patterns ranking subsequently. The categories of ’very high’ and ’high’ class collectively made up 37.1% and 26.3% of the total area, respectively. The flood vulnerability assessment of Pahang found that the Eastern, Southern, and central regions were at high risk of flooding due to intense precipitation, low-lying topography with steep inclines, proximity to the shoreline and rivers, and abundant flooded vegetation, crops, urban areas, bare ground, and rangeland. Conversely, areas with dense tree canopies or forests were less susceptible to flooding in this research area. The ROC analysis demonstrated strong performance on the validation datasets, with an AUC value of >0.73 and accuracy scores exceeding 0.71. Research on flood susceptibility mapping can enhance risk reduction strategies and improve flood management in vulnerable areas. Technological advancements and expertise provide opportunities for more sophisticated methods, leading to better prepared and resilient communities.</p></div>
Cited by 0Open access
Patrick P. K. Chan, Qinqin Zhang, Wing W. Y. Ng et al. · 2011
Multiple Classifier Systems (MCSs) are a method combining decisions of base classifiers. The set of the base classifiers is fixed in traditional MCSs. When applying MCSs in online learning environment, the base classifiers have to be updated frequently to adapt the change of the environment. However, updating classifiers is time consumed, especially when the number of base classifier is big. Therefore, a selection method with dynamic base classifier pool is proposed in this paper. Rather than updating the existing base classifiers, a new base classifier is added to MCSs. The new base classifier is trained by using the samples which far away from the training set. Experimental results show that that the proposed method outperforms the MCSs with the fix base classifier pool in term of accuracy.
Cited by 3
Dymitr Ruta A, Bogdan Gabrys B, Ipswich Ip Re · 2004
Individual classification models are recently challenged by combined pattern recognition systems, which often show better performance. In such systems the optimal set of classifiers is first selected and then combined by a specific fusion method. For a small number of classifiers optimal ensembles can be found exhaustively, but the burden of exponential complexity of such search limits its practical applicability for larger systems. As a result, simpler search algorithms and/or selection criteria are needed to reduce the complexity. This work provides a revision of the classifier selection methodology and evaluates the practical applicability of diversity measures in the context of combining classifiers by majority voting. A number of search algorithms are proposed and adjusted to work properly with a number of selection criteria including majority voting error and various diversity measures. Extensive experiments carried out with 15 classifiers on 27 datasets indicate inappropriateness of diversity measures used as selection criteria in favour of the direct combiner error based search. Furthermore, the results prompted a novel design of multiple classifier systems in which selection and fusion are recurrently applied to a population of best combinations of classifiers rather than the individual best. The improvement of the generalisation performance of such system is demonstrated experimentally. Ó 2004 Elsevier B.V. All rights reserved.
Cited by 0
Dymitr Ruta, Bogdan Gabryś · Information Fusion · 2004
Cited by 621
Arias Vicente, Marta, Arratia Quesada, Argimiro Alejandro, Duarte López, Ariel · RECERCAT (Consorci de Serveis Universitaris de Catalunya)
This work presents a content-based recommender system for machine learning classifier algorithms. Given a new data set, a recommendation of what classifier is likely to perform best is made based on classifier performance over similar known data sets. This similarity is measured according to a data set characterization that includes several state-of-the-art metrics taking into account physical structure, statistics, and information theory. A novelty with respect to prior work is the use of a robust approach based on permutation tests to directly assess whether a given learning algorithm is able to exploit the attributes in a data set to predict class labels, and compare it to the more commonly used F-score metric for evaluating classifier performance. To evaluate our approach, we have conducted an extensive experimentation including 8 of the main machine learning classification methods with varying configurations and 65 binary data sets, leading to over 2331 experiments. Our results show that using the information from the permutation test clearly improves the quality of the recommendations.
Cited by 0Open access
Cuiqin Hou, Yingju Xia, Zhuoran Xu et al. · 2016
Classifier competence is critical important for classifier ensemble. This study proposes an optimization problem on the neighborhood graph of data and develops an iteration algorithm to learn the competences of classifiers. The learned competences of classifiers not just reflect the competitiveness of classifiers, but also vary smooth on the neighboring data. Experimental results on five different data sets show the dynamic classifier selection based classification systems with the learned classifier competence perform competitively.
Cited by 7
Dunja Mladenić, Janez Brank, Marko Grobelnik et al. · 2004
This paper explores feature scoring and selection based on weights from linear classification models. It investigates how these methods combine with various learning models. Our comparative analysis includes three learning algorithms: Naïve Bayes, Perceptron, and Support Vector Machines (SVM) in combination with three feature weighting methods: Odds Ratio, Information Gain, and weights from linear models, the linear SVM and Perceptron. Experiments show that feature selection using weights from linear SVMs yields better classification performance than other feature weighting methods when combined with the three explored learning algorithms. The results support the conjecture that it is the sophistication of the feature weighting method rather than its apparent compatibility with the learning algorithm that improves classification performance.
Cited by 183
Sneha Aenugu, Lee Spector · Proceedings of the Genetic and Evolutionary Computation Conference · 2019
The lexicase parent selection method selects parents by considering performance on individual data points in random order instead of using a fitness function based on an aggregated data accuracy. While the method has demonstrated promise in genetic programming and more recently in genetic algorithms, its applications in other forms of evolutionary machine learning have not been explored. In this paper, we investigate the use of lexicase parent selection in Learning Classifier Systems (LCS) and study its effect on classification problems in a supervised setting. We further introduce a new variant of lexicase selection, called batch-lexicase selection, which allows for the tuning of selection pressure. We compare the two lexicase selection methods with tournament and fitness proportionate selection methods on binary classification problems. We show that batch-lexicase selection results in the creation of more generic rules which is favorable for generalization on future data. We further show that batch-lexicase selection results in better generalization in situations of partial or missing data.
Cited by 37Open access
David Windridge, Josef Kittler · Lecture notes in computer science · 2000
Cited by 5
Binbin Pan, Wen-Sheng Chen, Liping Deng et al. · Neural Computing and Applications · 2023
Cited by 2
Maria Muntean, Honoriu Vălean, Remus Joldes et al. · 2013
Abstract. Most of the time a lot of data means better results. This case is not valid all the time because sometimes we have a lot of redundant data and a lot of attributes that are weakly related to what we are trying to find out by evaluating the data. The main idea behind feature selection is to keep the data that bring the most amount of information for learning how to evaluate future data that are going to be fed to the system and to discard the features that do not bring any new information. In this paper we investigated whether feature selection methods can improve the accuracy and the time spent for classification. We have used the Support Vector Machine, a powerful classification technique based on kernels, which has proven to be efficient for nonlinearly separable input data. Basically what data mining tell us is that the more features we have, the better it is to make more accurate predictions about future instances. Practically the amount of training data is limited, and excessive features are going to slow down
Cited by 0
Paulo Henrique Gonçalves Leonel da [UNESP] Silva · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 2020
Outro
Cited by 0
Sotiris Kotsiantis, Panagiotis Pintelas · 2004
Feature subset selection is the process of identifying and removing from a training data set as much irrelevant and redundant features as possible. This reduces the dimensionality of the data and may allow learning algorithms to operate faster and more effectively. In some cases, accuracy on classification can be improved; in others, the result is a more compact, easily interpreted representation of the target concept. This paper presents a benchmarking comparison of five well-known wrapper feature selection methods. Experimental results are reported using four well known representative supervised learning algorithms. Finally, we use the produced conclusions for improving an educational software support tool. This tool is used to predict students ’ performance at Hellenic Open University based on students ’ demographic characteristics, their participation in face-to-face meetings with tutors and their marks in a few written assignments. I.
Cited by 1
Matthias Reif, Faisal Shafait, Markus Goldstein et al. · Pattern Analysis and Applications · 2012
Cited by 139
Lou Xin-yuan · 2008
In this paper,we delve into the characteristic of the stocks that Securities Analyst recommend by event-study.We try to find out these stocks whose characteristic accord with Securities Analyst recommend by naive Bayesian classifier.Given the information of stocks trading on the Shanghai Exchange A,we use naive Bayesian classifiers to classifing stocks and we can get recome outperform the market by choosing appropriate parameter of naive Bayesian classifier.The simulation results show that the proposed method is effective.
Cited by 0
Yan Yan, Xu-Cheng Yin, Zhibin Wang et al. · 2013
In ensemble learning, a higher accuracy can be achieved by integrating some classifiers instead of all the classifiers. But, it is very difficult to select the best classifier combination which can be seen as an optimization problem, from a pool of classifiers. To deal with this problem, we propose a new classifier selection method, Sorting-based Dynamic Classifier Ensemble Selection (SDES), which consists of two stages: (1) classifier sorting, and (2) dynamic ensemble selection on sorted classifier sequence. In the first stage, classifiers are sorted based on diversity, to avoid searching for the nearest neighbors in dynamic ensemble selection methods and greatly improve the selection efficiency. In the second stage, the optimal subset of classifiers is selected from the sorted classifier sequence based on confidence of test samples, to guarantee high accuracy of the optimal classifier subset. Experimental results have shown the effectiveness and high efficiency of the proposed method.
Cited by 6
Guangtao Wang, Xiaomei Yang, Xiaoyan Zhu · Lecture notes in computer science · 2016
Cited by 1
Rose Maria Valdovinos, Maribel Sanchez, Issachar Ruiz · Advances in intelligent and soft computing · 2009
Cited by 0
Sato Kudo · 2007
. The nonlinear classifier is effective for many practical problems. We have already proposed a method for constructing a nonlinear classifier using Legendre polynomials and have obtained good results on many actual data. In this approach, a set of original features is first extended to a large number of new features in a nonlinear fashion and then some substantial features are chosen for the nonlinear classifier. In this study, we have improved the selection process in the second stage by using some conventional feature selection algorithms. In addition, important features were selected from the original features in the preprocessing stage. The reduction in the number of the original features permits the nonlinear classifier to use a higher degree of polynomials. 1 Introduction In our previous study [1], we proposed a nonlinear classifier and confirmed its effectiveness on many data. In this method, a large number of Legendre polynomials of the original features are constructed as ex...
Cited by 0