Noisy Class Research Articles

Classification Performance Analysis of Decision Tree-Based Algorithms with Noisy Class Variable

Class noise is a common issue that affects the performance of classification techniques on real-world data sets. Class noise appears when a class variable in data sets has incorrect class labels. In the case of noisy data, the robustness of classification techniques against noise could be more important than the performance results on noise-free data sets. The decision tree method is one of the most popular techniques for classification tasks. The C4.5, CART, and random forest (RF) algorithms are considered to be three of the most used algorithms in decision trees. The aim of this paper is to reach conclusions on which decision tree algorithm is better to use for building decision trees in terms of its performance and robustness against class noise. In order to achieve this aim, we study and compare the performance of the models when applied to class variables with noise. The results obtained indicate that the RF algorithm is more robust to data sets with noisy class variable than other algorithms.

The Role of Morphology and Heart Rate Variability Features in Detecting Arrhythmias from Short ECGs

Detecting heart arrhythmias from short Electrocardiogram (ECG) recordings remains challenging since recordings are short and contaminated by noise. ECG morphology features and Heart Rate Variability (HRV) time and frequency domain features are widely used for classifying short ECG recordings. Here we investigate the relative roles of ECG morphology features and HRV time and frequency domain features in classifying short ECG recordings provided by the 2017 PhysioNet/Computing in Cardiology Challenge. The classification is performed separately by four machine learning models: Logistic Regression, Decision Tree, K Nearest Neighbors, and Convolutional Neural Network (CNN). Our best classification score is obtained using the deep learning 1-dimensional CNN model trained on HRV time domain features combined with ECG morphology features. It gives an overall F1 score of 0.70 and 0.73 for the cross validation and hidden test respectively when considering the average classification performance over all 4 categories: Atrial Fibrillation (AF), normal, other arrhythmias, and noisy signal. We found that HRV time domain features play an important role in detecting AF, normal, and other classes, whereas ECG morphology features play a key role in detecting the noisy class. When HRV frequency domain features are combined with HRV time domain features, they do not improve and often degrade classifications of short ECG recordings compared to classifications using only HRV time domain features. Combining ECG morphology features with HRV time domain features leads to a better classification performance for short ECG recordings. Feature-based deep learning could serve as a viable and less expensive approach for ECG classifications.

Task-Specific Loss for Robust Instance Segmentation With Noisy Class Labels

Deep learning methods have achieved significant progress in the presence of correctly annotated datasets in instance segmentation. However, object classes in large-scale datasets are sometimes ambiguous, which easily causes confusion. Besides, limited experience and knowledge of annotators can lead to mislabeled object semantic classes. To solve this issue, a novel method is proposed in this paper, which considers different roles of noisy class labels in different sub-tasks. Our method is based on two basic observations: firstly, the foreground-background annotation of a sample is correct even though its class label is noisy. Secondly, symmetric loss benefits the model robustness to noisy labels but harms the learning of hard samples, while cross entropy loss is the opposite. Based on the two basic observations, in the foreground-background sub-task, cross entropy loss is used to fully exploit correct gradient guidance. In the foreground-instance sub-task, symmetric loss is used to prevent incorrect gradient guidance provided by noisy class labels. Furthermore, we apply contrastive self-supervised loss to update features of all foreground, to compensate for insufficient guidance provided by partially correct labels especially in the highly noisy setting. Extensive experiments conducted with three popular datasets (i.e., Pascal VOC, Cityscapes and COCO) have demonstrated the effectiveness of our method in a wide range of noisy class label scenarios.

A hybrid approach for noise reduction-based optimal classifier using genetic algorithm: A case study in plant disease prediction

Plant diseases can cause significant losses to agricultural productivity; therefore, their early prediction is much needed. So far, many machine learning-based plant disease prediction models have been recommended, but these models face a problem of noisy class label dataset that degrades the performance. Noisy class label dataset results from the improper assignment of positive class labels into negative class data samples or vice versa. Hence, a precise and noise-free plant disease model is required for a better prediction. The current study proposes noise reduction-based hybridized classifiers for plant disease prediction. One tomato and four soybean disease datasets have been selected to conduct the proposed research. The Adaptive Sampling-based Class Label Noise Reduction (AS-CLNR) method has been used along with the Support Vector Machine (SVM) approach for noise reduction. The noise-minimized datasets have been fed into the Extreme Learning Machine (ELM), Decision Tree (DT), and Random Forest (RF) classifiers whose parameters are optimized using Genetic Algorithm (GA) for developing plant disease prediction models. The performances of all these models viz. Hybrid SVM-GA-ELM, Hybrid SVM-GA-DT, and Hybrid SVM-GA-RF have been evaluated using Accuracy, Area under ROC Curve, and F1-Score metrics. Further, these classifiers have been ranked using the statistical Friedman Test in which the Hybrid SVM-GA-RF classifier performed the best. Lastly, the Nemenyi test has also been performed to find out if significant differences exist between various classifiers or not. It was found that 33.33% of the total pairs of hybrid classifiers show a remarkably different performance from one another.

Self-Augmentation Based on Noise-Robust Probabilistic Model for Noisy Labels

Learning deep neural networks from noisy labels is challenging, because high-capacity networks attempt to describe data even with noisy class labels. In this study, we propose a self-augmentation method without additional parameters, which handles noisy labeled data based on small-loss criteria. To this end, we use small-loss samples by introducing a noise-robust probabilistic model based on a Gaussian mixture model (GMM), in which small-loss samples follow class-conditional Gaussian distributions. With this sample augmentation using the GMM-based probabilistic model, we can effectively solve over-parameterization problems induced by label inconsistency in small-loss samples. We further enhance the quality of the small-loss samples using our data-adaptive selection strategy. Consequently, our method prevents networks from over-parameterization and enhances their generalization performance. Experimental results demonstrate that our method outperforms state-of-the-art methods for learning with noisy labels on several benchmark datasets. The proposed method produced a remarkable performance gap of up to 12% compared with the previous state-of-the-art methods on CIFAR dataset.

Knowledge discovery from noisy imbalanced and incomplete binary class data

Class imbalance creates a considerable impact on the classification of instances using traditional classifiers. Class imbalance, along with other difficulties, creates a significant impact on recognizing instances of minority class. Researchers work in various directions to mitigate class imbalance effect along with noise as well as missing values in datasets. However, combined studies of noisy class imbalance along with incomplete datasets have not been performed yet. This article contains a detailed analysis of 84 different machine learning models to deal with noisy binary class imbalanced and incomplete data using AUC, G-Mean, and F1-score as performance metrics. This article contains a detailed experiment considering missing value imputation and oversampling techniques. The article contains three comparisons: first missing value imputation techniques in incomplete and binary class imbalanced data, second, resampling techniques in noisy binary class imbalanced data, and third, combined techniques in noisy binary class imbalanced and incomplete data. We conclude that MICE and KNN techniques perform well with an increase in the imbalanced dataset's missing value from the first comparison. In second comparison, the SMOTE-ENN technique performs better than state-of-art in noisy binary class imbalanced datasets, and in the third comparison, we conclude that MICE with SMOTE-ENN technique perform well compared to the rest of the techniques.

Improved Hybrid Bag-Boost Ensemble With K-Means-SMOTE–ENN Technique for Handling Noisy Class Imbalanced Data

Abstract A class imbalance problem plays a vital role while dealing with classes with rare number of instances. Noisy class imbalanced datasets create considerable effect on the machine learning classification of classes. Data resampling techniques commonly used for handling class imbalance problem show insignificant behavior in noisy imbalanced datasets. To cure curse of data resampling technique in noisy class imbalanced data, we have proposed improved hybrid bag-boost with proposed resampling technique model. This model contains proposed resampling technique used for handling noisy imbalanced datasets. Proposed resampling technique comprises K-Means SMOTE (Synthetic Minority Oversampling TEchnique) as an oversampling technique and edited nearest neighbor (ENN) undersampling technique used as noise removal. This resampling technique is used to mitigate noise in imbalanced datasets at three levels, i.e. first clusters datasets using K-Means clustering technique, SMOTE inside clusters for handling imbalance by inducing synthetic instances of class in minority and lastly, using ENN technique to remove instances that create noise afterwards. Experiments were performed using 11 binary imbalanced datasets by varying attribute noise percentages, and by using area under receiver operating curve as performance metrics. Experimental results confirmed that proposed model shows better results than the rest. Moreover, it is also confirmed that proposed technique performs better with an increased noise percentage in binary imbalanced datasets.

Scalable Zero-Shot Learning via Binary Visual-Semantic Embeddings.

Zero-shot learning aims to classify visual instances from unseen classes in the absence of training examples. This is typically achieved by directly mapping visual features to a semantic embedding space of classes (e.g., attributes or word vectors), where the similarity between the two modalities can be readily measured. However, the semantic space may not be reliable for recognition due to the noisy class embeddings or visual bias problem. In this work, we propose a novel Binary embedding based Zero-Shot Learning (BZSL) method, which recognizes visual instances from unseen classes through an intermediate discriminative Hamming space. Specifically, BZSL jointly learns two binary coding functions to encode both visual instances and class embeddings into the Hamming space, which well alleviates the visual-semantic bias problem. As a desiring property, classifying an unseen instance thereby can be efficiently done by retrieving its nearest-class codes with minimal Hamming distance. During training, by introducing two auxiliary variables for the coding functions, we formulate an equivalent correlation maximization problem, which admits an analytical solution. The resulting algorithm thus enjoys both highly efficient training and scalable novel class inferring. Extensive experiments on four benchmark datasets, including the full ImageNet Fall 2011 dataset with over 20K unseen classes, demonstrate the superiority of our method on the zero-shot learning task. Particularly, we show that increasing the binary embedding dimension can inevitably improve the recognition accuracy.

The Implementation of Jigsaw Technique and Student Team Achievement Division (STAD) in Teaching Reading

The objectives of the research were aimed at finding out the responses the students had during the implementation of Jigsaw technique and Student Team Achievement Division (STAD). This research was done through a qualitative design. The population of this research was the second grade students at SMPN 1 Abung Surakarta. The result showed that there were some obstacles found during the implementation of the Jigsaw and STAD Technique as follows: (1) limited vocabulary knowledge; (2) difficulty level of the text (readability); (3) noisy class situation; (4) no students background knowledge; (5) hesitancy to express opinions; (6) class dominated by the smart students. The results showed that there were obstacles that students faced during teaching, and learning reading comprehension in the class using Jigsaw, and STAD Technique. Keywords: Jigsaw Technique, reading comprehension, STAD Technique DOI: http://dx.doi.org/10.23960/aksara/v19i1.pp28-36

Jigsaw and Student Team Achievement Division (STAD) Techniques in Teaching English Reading

The objectives of the research were aimed at finding out the responses the students had during the implementation of Jigsaw technique and Student Team Achievement Division (STAD). This research was done through a qualitative design. The population of this research was the second grade students at SMPN 1 Abung Surakarta. The result showed that there were some obstacles found during the implementation of the Jigsaw and STAD Technique as follows: (1) limited vocabulary knowledge; (2) difficulty level of the text (readability); (3) noisy class situation; (4) no students background knowledge; (5) hesitancy to express opinions; (6) class dominated by the smart students. The results showed that there were obstacles that students faced during teaching, and learning reading comprehension in the class using Jigsaw, and STAD Technique. Keywords: Jigsaw Technique, reading comprehension, STAD Technique DOI: http://dx.doi.org/10.23960/aksara/v18i2.pp169-177

The factors associated with hearing impairment in school children

Background: The morbidity of ENT infections have increased considerably in recent decades, so that they are now amongst the most common chronic disorders of childhood. Awareness of hearing deficit is poor in India. The major proportion of childhood hearing impairment in developing countries is secondary to preventable causes. At the same time, even mild degrees of hearing impairment can affect proper learning in noisy class rooms where speech is produced at a distance. This in turn can have significant impact on the scholastic performance and overall development.Methods: A cross sectional study was conducted to identify the factors leading to hearing impairment in school children aged 8years to 14 years in the schools in Vadamavanthal area during the months of January 2013 to July 2013. The students were interviewed through pre - tested questionnaire and aural examination was done by a qualified otolaryngologist.Results: Out of 700 school, 216(30.9%) school children were suffering from hearing impairment. 139 (19.9%) students were found to be suffering from impacted wax, 67 (9.6%) were suffering from otitis media with effusion, 17 (2.4%) students were suffering from chronic suppurative otitis media and 5 (0.7%) students were suffering from foreign body impaction in ear.Conclusions: Simple measures like regular screening for ear diseases and hearing assessment done at the school level can help identify hearing impairment which can be modified with early health education for students and teachers.

Evolutionary Learning of Concepts

Concept learning is a kind of classification task that has interesting practical applications in several areas. In this paper, a new evolutionary concept learning algorithm is proposed and a corresponding learning system, called ECL (Evolutionary Concept Learner), is implemented. This system is compared to three traditional learning systems: MLP (Multilayer Perceptron), ID3 (Iterative Dichotomiser) and NB (Naïve Bayes). The comparison takes into account target concepts of varying complexities (e.g., with interacting attributes) and different qualities of training sets (e.g., with imbalanced classes and noisy class labels). The comparison results show that, although no single system is the best in all situations, the proposed system ECL has a very good overall performance.

The strong type of silent

No-one wants to have ‘the noisy class’, yet silence can be an anathema in today's classrooms and training. Do not miss out on this valuable tool, says Jackie Cosh.

Private and quantum capacities of more capable and less noisy quantum channels

Two classes of quantum channels, which we call more capable and less noisy, are introduced. The more capable class consists of channels such that the quantum capacities of the complementary channels to the environments are zero. The less noisy class consists of channels such that the private capacities of the complementary channels to the environment are zero. For the more capable class, it is clarified that the private capacity and quantum capacity coincide. For the less noisy class, it is clarified that the private capacity and quantum capacity can be single letter characterized.

The New Teaching Method Using a Wireless Microphone for Hard-of-Hearing Children

The author reported a new teaching method for hard-of-hearing children. Teachers voice are transmitted through a wireless microphone to the receiver of FM transistor radio weared by the hard-of-hearing child.As a point of advantage of this method the child experiences lesser fatigue comparing with the hearing aid, especially in the noisy class room of the lower elementary grades.This is a convenient method which can be employed in the ordinary school in the district where neither day nor residential schools for hard-of-hearing children are founded.

THE INVESTIGATION OF THE NOISE-SURROUNDINGS AT THE PRIMARY SCHOOL ROOMS IN KYOTO

It is the object of this investigation to examine the noise-surroundings in the pri- mary school rooms and contribute to the improvement of them. For this object, not only the intensity and constitution of the noise in each school room were examined by using the sound level meter but also the degree of the inconvenience caused by noise were demonstrated clearly on the results of the articulation test and the following re- sults were obtained. 1) The most of the noises in the class rooms of the primary school originate from the means of traffic and not from the industrial noises in Kyoto. 2) There are very few class rooms of the primary school in which the average level of the noise is under 40 phon and even in such class rooms the articulation deteriora- tes already from 10 to 20% than in the sound proof room, especially in the extremely noisy class rooms the deterioration of the articulation reaches to about 50%. 3) The degree of the inconvenience in each class rooms caused by noise that was indicated on the results on the articulation test is in proportion to the intensity of the average noise level, and when the intensity of the average noise level is almostequal to the degree of the inconvenience, the degree of the in convenience in each class room is controled by the deviatian range of the noise level. 4) When the articulation test was carried out using logatom , the allowable lowest level of the articulation in the class rooms of the primary school corresponds to the articulation of the adults in the sound proof room who have conductive deaffness showing 30 db loss by pure tone audiometry, and from this point of view, it seems pro- per to decide that the allowable average noise level in the class rooms should not -exceed 45 phon.