China
Africa
Explore chapters and articles related to this topic
Machine learning classifier for fault classification in photovoltaic system
Published in Rajesh Singh, Anita Gehlot, Intelligent Circuits and Systems, 2021
V.S. Bharath Kurukuru, Mohammed Ali Khan, Ahteshamul Haque, Arun Kumar Tripathi
The methodology adopted for fault classification is structured as depicted in Figure 5.1. In this research, the fault classification technique is used to learn a model called a classifier. Data linking to various faults and normal operational conditions of the PV system is bifurcated into two groups; testing and training set. During the training phase, the training set of data is fed to the classifier for labeled data set into one of the classes depending on the target output. Fivefold cross-validation is applied to validate the trained data. During the testing phase, depending upon the target output test sample is verified. Once essential features have been identified, the classification of a fault condition is straightforwardly performed. The SVM technique is used in this work for fault classification due to its advantages with nonlinear data classification as mentioned in the section above. For simulation evaluation, the values of input data (i.e. the extracted features) are tabulated and imported to the classification program.
Modern machine learning techniques and their applications
Published in Amir Hussain, Mirjana Ivanovic, Electronics, Communications and Networks IV, 2015
Mirjana Ivanović, Miloš Radovanovic
Overfitting is a related notion to the bias-variance tradeoff within supervised learning, and refers to the notion that a classifier can be trained "too much," in the sense of maximizing its performance on the training set, which may in fact lead to suboptimal performance on a separate test set and real life data. Overfitting may come as a consequence of a small or large number of training instances, noisy data, and/or high dimensionality. Some classifiers are more prone to overfitting than others, and many of them employ complex strategies to avoid it. The philosophical equivalent of the problem lies in the Occam's razor principle, which in ML terms translates to preferring a simple model which reasonably fits the data, to a complex one which does so more accurately.
Fundamental Pattern Recognition Concepts
Published in Manas Kamal Bhuyan, Computer Vision and Image Processing, 2019
Once a classifier is designed via training, it is necessary to check the performance of the designed system. This is called “testing.” Testing means submitting the test patterns to the designed classifier and then evaluating the performance of the system. Generally, the classification accuracy of the system is taken as the measure for its performance that is defined as: AccuracyRate=TotalnumberofcorrectlyrecognizedtestpatternsTotalnumberoftestpatterns
An improved MI recognition by localising feature extraction in both frequency and time domains
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2023
M. K. M. Rahman, H. M. Tanvir Shuvo
Final important step of identifying an MI pattern is classification. Every classification algorithm contains two steps: namely, training and testing phase. A classifier is trained with a training dataset during which parameters of the classifier are optimised through a recursive procedure using both the class labels and the extracted features from the training dataset. Once the training is complete, the classifier can predict the class of an unknown EEG signal by using its features. To process an unknown data, a raw EEG signal is preprocessed, divided into frequency bands, spatial filters are applied, and finally time-localised features are extracted. Using the feature, the trained classifier then identifies the MI signal. The LDA classifier is widely utilised in the BCI sector because of its consistent performance (Lotte et al. 2007) and minimal computing demand. In this study, LDA is used by default. However, SVM and ANN are also used for comparison.
COVID-19 lung infection detection using deep learning with transfer learning and ResNet101 features extraction and selection
Published in Waves in Random and Complex Media, 2022
Raja Nadir Mahmood Khan, Lal Hussain, Ala Saleh Alluhaidan, Abdul Majid, Kashif J. Lone, Rufat Verdiyev, Fahd N. Al-Wesabi, Tim Q. Duong
In an effort to further improve the accuracy of classifying COVID-19 lung infection, we utilized the help of multiple supervised machine learning (ML) classifiers and feature selection methods. A classifier is an algorithm that connects the input data to a distinct category. To illustrate the three examples of ML classifiers that were used in this study, K-nearest neighbor (KNN), eXtreme gradient boosting tree (XGB-T) and support vector machine radial base (SVM-R) are presented. KNN is a classifier that is used when searching for similarities. The best first and depth first algorithms are used to find KNNs and to search hierarchy that contains data [4]. XGB-T, another example of a classifier we employed in this study, is mainly used for tabular data or structured data which are implemented using gradient boosted decision trees designed for performance and speed [5]. SVM-R, the third classifier applied in this study, is a SVM programed in R language. SVM-R outputs an optimal hyperplane on new examples [5].
Detection of AAC compression using MDCT-based features and supervised learning
Published in Journal of Experimental & Theoretical Artificial Intelligence, 2022
José Juan García-Hernández, Wilfrido Gómez-Flores
The procedure shown in Figure 1 builds a feature vector containing the variance of MDCT coefficients of an input audio signal. Thus, a high-dimensional feature space is generated, which is typical in spectrum analysis (Verleysen & François, 2005). When classifying high-dimensional data, it is necessary to deal with sparseness and closeness, known as ‘the curse of dimensionality’. Data sparseness occurs when the volume of the feature space overgrows so that the data cannot keep up and becomes sparse. On the other hand, data closeness refers to the similarity of data points as a function of dimensionality. In high-dimensional spaces, data points are approximately equidistant from each other (Sammut & Webb, 2017). As a result, the classifier learns the appearance of specific instances and exceptions of the training dataset; thus, the resulting classifier would have poor generalisation on new data, which is called overfitting (Liu & Gillies, 2016).