Anomaly detection has been widely studied in many domains where abnormal samples with more valuable information than normal samples are of interest to researchers. Numerous methods have been developed for detecting anomalies, with one-class classification techniques being especially popular due to their ability to train using only normal samples. However, the presence of anomalies in the unlabeled data may negatively impact the accuracy of the model. Although some studies have attempted to improve robustness by removing potential anomalies or assigning lower weights to them during data preprocessing, these approaches are limited in effectiveness. This limitation arises because sample preprocessing is conducted independently of anomaly scoring, leading to the potential misclassification of anomalies as normal samples or their assignment of high weights, ultimately resulting in suboptimal detection performance. In this paper, an ensemble-based joint training framework is proposed for anomaly detection, called E-ORAD, to enhance robustness. E-ORAD jointly optimizes sample weights and anomaly scores within a unified ensemble model. Specifically, to compute the weight of each sample, a prior knowledge-based regularization term is incorporated into the objective function. Additionally, an anomaly score-based hinge loss function is developed to ensure that abnormal samples are assigned higher scores than normal samples. An alternating iteration method is then introduced to solve the proposed model. Experimental results on 10 real-world datasets demonstrate that E-ORAD outperforms other state-of-the-art methods in terms of detection performance.

Early breast cancer diagnosis is crucial for improving treatment outcomes for women. Addressing class imbalance in breast cancer data is essential for enhancing detection accuracy, yet traditional machine learning methods often overlook this imbalance, limiting their performance. To tackle this issue, we propose a robust enhanced ensemble learning method (REEL). Specifically, a double-level over-sampling technology is developed to increase the diversity of synthesized minority breast cancer samples before model training, and an improved Random Forest is proposed to reconcile the bias and variance. In addition, a data-driven based particle swarm optimization algorithm automatically is used to select the value of parameters for base classifiers. Experimental results on breast cancer datasets and 19 other imbalanced datasets validate that our method outperforms other algorithms in terms of accuracy, F1 score, and AUC.These findings confirm that our method can further improve classification accuracy and has significant application value in the diagnosis of breast cancer.

Cropland non-grain and non-agriculturalization monitoring (CNNM) is of great significance for ensuring national food security. Recently, semantic segmentation has been a promising approach that can fully exploit the advantages of high-resolution remote sensing imagery. It is of great significance to clarify the performance of semantic segmentation algorithms for CNNM and their influencing factors. This paper contributes to research on deep learning-based semantic segmentation models for CNNM by investigating the robustness and generalization ability of detection models in complex scenarios. To investigate these models, seven mainstream segmentation models were examined with respect to two self-constructed unmanned aerial vehicle (UAV)-based cropland monitoring datasets under various experimental settings. The experimental results reveal several key findings: For the datasets employed in the present study, remote sensing semantic segmentation models demonstrate superior performance relative to generic visual models. Increasing backbone depth does not guarantee significant performance improvements. The recognition challenges posed by task-specific categories such as agricultural facility land further underscore the need for tailored deep learning semantic segmentation algorithms for CNNM. These experimental results provide valuable insights into the practical performance of deep learning semantic segmentation models and offer useful guidance for future semantic segmentation research in the field of CNNM.