Tackling label noise with multi-class decomposition using fuzzy one-class support vector machines

Noise correction to improve data and model quality for crowdsourcing

Supervised classification systems used for land cover mapping require accurate reference databases. These reference data come generally from different sources such as field measurements, thematic maps, or aerial photographs. Due to misregistration, update delay, or land cover complexity, they may contain class label noise, i.e., a wrong label assignment. This study aims at evaluating the impact of mislabeled training data on classification performances for land cover mapping. Particularly, it addresses the random and systematic label noise problem for the classification of high resolution satellite image time series. Experiments are carried out on synthetic and real datasets with two traditional classifiers: Support Vector Machines (SVM) and Random Forests (RF). A synthetic dataset has been designed for this study, simulating vegetation profiles over one year. The real dataset is composed of Landsat-8 and SPOT-4 images acquired during one year in the south of France. The results show that both classifiers are little influenced for low random noise levels up to 25%–30%, but their performances drop down for higher noise levels. Different classification configurations are tested by increasing the number of classes, using different input feature vectors, and changing the number of training instances. Algorithm complexities are also analyzed. The RF classifier achieves high robustness to random and systematic label noise for all the tested configurations; whereas the SVM classifier is more sensitive to the kernel choice and to the input feature vectors. Finally, this work reveals that the cross-validation procedure is impacted by the presence of class label noise.

Local ensemble learning from imbalanced and noisy data for word sense disambiguation

Label noise in medical image classification datasets significantly hampers the training of supervised deep learning methods, undermining their generalizability. The test performance of a model tends to decrease as the label noise rate increases. Over recent years, several methods have been proposed to mitigate the impact of label noise in medical image classification and enhance the robustness of the model. Predominantly, these works have employed CNN-based architectures as the backbone of their classifiers for feature extraction. However, in recent years, Vision Transformer (ViT)-based backbones have replaced CNNs, demonstrating improved performance and a greater ability to learn more generalizable features, especially when the dataset is large. Nevertheless, no prior work has rigorously investigated how transformer-based backbones handle the impact of label noise in medical image classification. In this paper, we investigate the architectural robustness of ViT against label noise and compare it to that of CNNs. We use two medical image classification datasets-COVID-DU-Ex, and NCT-CRC-HE-100K-both corrupted by injecting label noise at various rates. Additionally, we show that pretraining is crucial for ensuring ViT's improved robustness against label noise in supervised training.

Objective. Learning to classify cardiac abnormalities requires large and high-quality labeled datasets, which is a challenge in medical applications. Small datasets from various sources are often aggregated to meet this requirement, resulting in a final dataset prone to label noise due to inter- and intra-observer variability and different expertise. It is well known that label noise can affect the performance and generalizability of the trained models. In this work, we explore the impact of label noise and self-learning label correction on the classification of cardiac abnormalities on large heterogeneous datasets of electrocardiogram (ECG) signals. Approach. A state-of-the-art self-learning multi-class label correction method for image classification is adapted to learn a multi-label classifier for electrocardiogram signals. We evaluated our performance using 5-fold cross-validation on the publicly available PhysioNet/Computing in Cardiology (CinC) 2021 Challenge data, with full and reduced sets of leads. Due to the unknown label noise in the testing set, we tested our approach on the MNIST dataset. We investigated the performance under different levels of structured label noise for both datasets. Main results. Under high levels of noise, the cross-validation results of self-learning label correction show an improvement of approximately 3% in the challenge score for the PhysioNet/CinC 2021 Challenge dataset and an improvement in accuracy of 5% and reduction of the expected calibration error of 0.03 for the MNIST dataset. We demonstrate that self-learning label correction can be used to effectively deal with the presence of unknown label noise, also when using a reduced number of ECG leads.

On the usefulness of one-class classifier ensembles for decomposition of multi-class problems

Breast cancer is widely prevalent in women. To diagnose it, pathologists evaluate breast histopathology images under a microscope in various magnifications. However, the number of pathologists per population is low in many countries, and a mistake is probable. Nowadays, deep learning has become a popular artificial intelligence research trends. Even though deep neural networks have obtained promising results in image processing, the need for extensive training data avoids its use in medical image processing. After the evaluation of various state-of-the-art deep learning methods and algorithms in medical data processing, this study proposes an effective deep transfer learning-based model, relying on pre-trained DCNN using an extensive data of ImageNet dataset which improves state-of the-art systems in both binary and multi-class classification. The weights of the pre-trained DesneNet121 on the Imagenet are transferred as initial weights first, and then the model is fine-tuned with a deep classifier along with data augmentation to distinguish various malignant and benign samples in the two categories of binary and multi-class classification. In the multi-class classification, the proposed model obtained up to 97% image-level accuracy. In binary classification, the model obtained up to 100% image-level accuracy. The achieved-results are outperforming previous studies accuracies in multiple performance metrics in breast cancer CAD systems. Moreover, the proposed method is both flexible and scalable, meaning it can be easily expanded to cover the detection of other types of diseases in the future and be integrated with more CNNs to increase its generalization capabilities.

A deep learning fusion network trained with clinical and high-frequency ultrasound images in the multi-classification of skin diseases in comparison with dermatologists: a prospective and multicenter study

As security threats become more complex, the need for effective intrusion detection systems (IDSs) has grown. Traditional machine learning methods are limited by the need for extensive feature engineering and data preprocessing. To overcome this, we propose two enhanced hybrid deep learning models, an autoencoder–convolutional neural network (Autoencoder–CNN) and a transformer–deep neural network (Transformer–DNN). The Autoencoder reshapes network traffic data, addressing class imbalance, and the CNN performs precise classification. The transformer component extracts contextual features, which the DNN uses for accurate classification. Our approach utilizes an enhanced hybrid adaptive synthetic sampling–synthetic minority oversampling technique (ADASYN-SMOTE) for binary classification and enhanced SMOTE for multi-class classification, along with edited nearest neighbors (ENN) for further class imbalance handling. The models were designed to minimize false positives and negatives, improve real-time detection, and identify zero-day attacks. Evaluations based on the CICIDS2017 dataset showed 99.90% accuracy for Autoencoder–CNN and 99.92% for Transformer–DNN in binary classification, and 99.95% and 99.96% in multi-class classification, respectively. On the NF-BoT-IoT-v2 dataset, the Autoencoder–CNN achieved 99.98% in binary classification and 97.95% in multi-class classification, while the Transformer–DNN reached 99.98% and 97.90%, respectively. These results demonstrate the superior performance of the proposed models compared with traditional methods for handling diverse network attacks.

Gas turbine combustor anomaly detection plays a critical role in reducing operation and maintenance costs in power plant operations. One-class classification, to learn a model that properly describes the normal samples, is one of many anomaly detection approaches that have shown promising performance in real-world applications. Extreme learning machines (ELMs), a recent developed machine learning technique, have outperformed SVMs and other machine learning methods for many machine learning problems. ELMs have been originally used for binary or multiclass classification and regression as well. Using ELM for one-class classification has emerged very recently, but has not been widely explored for real world applications. In this paper, we adopt ELMs to a new application domain - industrial machine condition monitoring. More specifically, we apply one-class ELMs for more accurate anomaly detection of gas turbine combustors. We evaluate different one-class ELMs and compare them against other one-class classifiers. Using a real-world gas turbine combustor anomaly detection as the case study, we demonstrate that one-class ELMs can be more effective than other one-class classification algorithms in early detecting gas turbine combustor anomalies.

Ultrawideband (UWB) radio frequency positioning technology has found extensive applications in indoor and outdoor localization due to its strong anti-interference capabilities, high penetration power, and precise measurement accuracy. However, its positioning accuracy significantly decreases under nonline-of-sight (NLOS) conditions, particularly in dynamic NLOS environments. Therefore, it is essential to identify NLOS propagation and mitigate its associated errors. To overcome these issues, this study proposes a novel enhanced domain-adversarial neural network for UWB signal occlusion recognition (EDANN-SOR). EDANN-SOR integrates hybrid manually extracted channel impulse response (CIR) features, effectively aligning the source and target domains and extracting discriminative and domain-invariant UWB CIR features in dynamic NLOS environments. Through adversarial learning, our method achieves high generalization with minimal CIR sample requirements, reducing the need for extensive data collection and significantly enhancing robustness in distance measurement and positioning under dynamic NLOS conditions. In contrast to state-of-the-art NLOS identification and transfer learning (TL) techniques, our method excels in both binary and multiclass NLOS classification. Specifically, it achieves an accuracy of over 97.36% in conventional LOS/NLOS binary classification across scenarios and 97.07% in multiclass classification. The integration of manually extracted features has been proved effective in improving the EDANN-SOR model’s ability to distinguish among LOS, human, and glass obstacles.

The softmax and binary classifier are commonly preferred for image classification applications. However, as softmax is specifically designed for categorical classification, it assumes each image has just one class label. This limits its applicability for problems where the number of labels does not equal one, most notably zero- and multi-label problems. In these challenging settings, binary classifiers are, in theory, better suited. However, as they ignore the correlation between classes, they are not as accurate and scalable in practice. In this paper, we start from the observation that the only difference between binary and softmax classifiers is their normalization function. Specifically, while the binary classifier self-normalizes its score, the softmax classifier combines the scores from all classes before normalisation. On the basis of this observation we introduce a normalization function that is learnable, constant, and shared between classes and data points. By doing so, we arrive at a new type of binary classifier that we coin quasibinary classifier. We show in a variety of image classification settings, and on several datasets, that quasibinary classifiers are considerably better in classification settings where regular binary and softmax classifiers suffer, including zero-label and multi-label classification. What is more, we show that quasibinary classifiers yield well-calibrated probabilities allowing for direct and reliable comparisons, not only between classes but also between data points.

The modern automotive system, based on the convergence of information and communication technologies, is equipped with various functions to ensure vehicle safety and convenience of the driver. A driver-identification technology is an effective method to perform vehicle-theft detection. It can also provide customized driver-personalization services, such as healthcare or insurance. In this article, we propose and evaluate a driver-identification method based on wavelet transform by performing driving-pattern analysis for each driver. We compare the performances of three different machine-learning algorithms, namely Support Vector Machine (SVM), Random Forest, and XGBoost for performing driver identification. The proposed method is applicable to both binary and multiclass classifications for the driving data of five drivers. In the case of motorway, the XGBoost classifier identifies each driver and delivers an accuracy of up to 96.18% in binary classification and an accuracy of 91.6% in multiclass classification. Moreover, in the case of an urban road, the SVM classifier achieves an accuracy of up to 95.07% in binary classification and accuracy of 89.06% in multiclass classification. The proposed method provides a context for a better understanding of the association between driver behavior, which is an in-vehicle event, and mechanical reactions. Our results shall help researchers to broaden the understanding of driver identification using in-vehicle data.

The field of histopathological image analysis has evolved significantly with the advent of digital pathology, leading to the development of automated models capable of classifying tissues and structures within diverse pathological images. Artificial intelligence algorithms, such as convolutional neural networks, have shown remarkable capabilities in pathology image analysis tasks, including tumor identification, metastasis detection, and patient prognosis assessment. However, traditional manual analysis methods have generally shown low accuracy in diagnosing colorectal cancer using histopathological images. This study investigates the use of AI in image classification and image analytics using histopathological images using the histogram of oriented gradients method. The study develops an AI-based architecture for image classification using histopathological images, aiming to achieve high performance with less complexity through specific parameters and layers. In this study, we investigate the complicated state of histopathological image classification, explicitly focusing on categorizing nine distinct tissue types. Our research used open-source multi-centered image datasets that included records of 100.000 non-overlapping images from 86 patients for training and 7180 non-overlapping images from 50 patients for testing. The study compares two distinct approaches, training artificial intelligence-based algorithms and manual machine learning models, to automate tissue classification. This research comprises two primary classification tasks: binary classification, distinguishing between normal and tumor tissues, and multi-classification, encompassing nine tissue types, including adipose, background, debris, stroma, lymphocytes, mucus, smooth muscle, normal colon mucosa, and tumor. Our findings show that artificial intelligence-based systems can achieve 0.91 and 0.97 accuracy in binary and multi-class classifications. In comparison, the histogram of directed gradient features and the Random Forest classifier achieved accuracy rates of 0.75 and 0.44 in binary and multi-class classifications, respectively. Our artificial intelligence-based methods are generalizable, allowing them to be integrated into histopathology diagnostics procedures and improve diagnostic accuracy and efficiency. The CNN model outperforms existing machine learning techniques, demonstrating its potential to improve the precision and effectiveness of histopathology image analysis. This research emphasizes the importance of maintaining data consistency and applying normalization methods during the data preparation stage for analysis. It particularly highlights the potential of artificial intelligence to assess histopathological images.

In many applications of classifier learning, training data suffers from label noise. Deep networks are learned using huge training data where the problem of noisy labels is particularly relevant. The current techniques proposed for learning deep networks under label noise focus on modifying the network architecture and on algorithms for estimating true labels from noisy labels. An alternate approach would be to look for loss functions that are inherently noise-tolerant. For binary classification there exist theoretical results on loss functions that are robust to label noise. In this paper, we provide some sufficient conditions on a loss function so that risk minimization under that loss function would be inherently tolerant to label noise for multiclass classification problems. These results generalize the existing results on noise-tolerant loss functions for binary classification. We study some of the widely used loss functions in deep networks and show that the loss function based on mean absolute value of error is inherently robust to label noise. Thus standard back propagation is enough to learn the true classifier even under label noise. Through experiments, we illustrate the robustness of risk minimization with such loss functions for learning neural networks.