Federated Learning (FL) trains shared models across distributed clients without pooling raw data, but its behaviour under heterogeneous data distributions and the associated privacy risks remain imperfectly understood. This paper presents a controlled empirical comparison of Federated Averaging (FedAvg), Federated Proximal (FedProx), Federated Adaptive Moment Estimation (FedAdam), and Differentially Private Federated Averaging (DP-FedAvg) under matched conditions. The experiments use CIFAR-10 and MNIST with Non-Independent and Identically Distributed (non-IID) client partitions, including Dirichlet and class-restricted partitioning. Each condition is repeated across 20 independent seeds. We examine how optimisation choice shapes convergence and stability, how differential privacy affects classification performance, and how privacy exposure changes over the course of training. We find that FedAdam improves performance under severe client heterogeneity, where client gradients conflict strongly. FedProx changes the training trajectory, but we do not find it to improve cross-seed stability. DP-FedAvg reduces the success of gradient-reconstruction and membership-inference attacks, but this protection comes with a measurable reduction in classification performance due to gradient clipping and additive noise. A key finding is that privacy exposure is temporally concentrated rather than uniform across training. It is highest in the earliest rounds, when gradients are largest and most informative, and declines as training progresses. These results suggest that private FL systems should account for training dynamics, with particular emphasis on protection during early training rounds.

In the aviation industry, inventory management based on non-smooth demand forecasting is an ongoing challenge. Generally, aircraft parts lack easily observable demand patterns. Through twenty-six articles published from 2012 to 2023, this literature review discusses the performance of several forecasting techniques using time series data. These methodologies include statistical forecasting methods, such as Croston’s method and single exponential smoothing (SES), and machine learning (ML) models built using ensembles of decision trees or neural networks. Research into demand forecasting does not only focus on achieving and maintaining accurate predictions, but also on lowering costs associated with over- or under-forecasting. While classical methods remain popular due to their ease of use, explainability, and cost-effectiveness, ML models demonstrate potential for higher accuracy under optimal conditions. With a focus on the aviation domain, this survey provides an extensive overview of demand forecasting methods that can assist businesses with improving inventory control of aircraft stock.

In this research, we assess the impact of training One-Class Classification (OCC) algorithms on the majority class versus the minority class, by employing highly imbalanced large and big datasets. It is worth highlighting that the accessibility of class data can be a major hurdle in training models. Depending on the situation, it may be possible to obtain one class within a reasonable time frame while not obtaining another. Our study centers around detecting fraudulent activities using the Credit Fraud Detection Dataset and a dataset derived from Medicare Part B, Part D, and the Durable Medical Equipment, Prosthetics, Orthotics and Supplies (DMEPOS) data, combined with the labels of the List of Excluded Individuals and Entities (LEIE) records. The Credit Card Fraud Detection Dataset, with its real-world transactional data and pronounced class disparity, serves as a benchmark for fraud detection. Notably, it stands as the only large-scale public dataset tailored for credit card fraud analysis. In Part B, Part D, and DMEPOS, researchers can investigate national trends and patterns based on Medicare big data. For our experiments, we use One-Class Gaussian Mixture Model (GMM), One-Class Adversarial Nets (OCAN), and One-Class Support Vector Machine (SVM). Their classification performance is evaluated with the Area Under the Precision-Recall Curve (AUPRC) and Area Under the Receiver Operating Characteristic Curve (AUC). Given the challenges posed by imbalanced datasets, our results indicate that strategically focusing training on the majority class yields better results.

One consequence of the widespread use of (IoT) devices is an increase in the volume of attacks on (IoT) networks. In this study, we focus on the Bot-IoT dataset, with the aim of classifying its four types of attacks: Denial-of-Service (DoS), Distributed Denial-of-Service (DDoS), Reconnaissance, and Information Theft. Our contribution is based on the evaluation of the Random Undersampling (RUS) technique and ensemble Feature Selection Techniques (FSTs). Our results indicate that RUS has a positive impact on overall classification performance. Furthermore, our results show that the FSTs are beneficial for DoS, Reconnaissance, and Information Theft classification but not for DDoS classification. Finally, we note that the ensemble classifiers have generally outperformed the nonensemble classifiers in our study.

This paper surveys the latest unsupervised anomaly detection methodologies applied to health insurance fraud, covering studies from 2017 to 2024. Our review includes a variety of machine-learning approaches, evaluating their effectiveness in handling complex, high-dimensional, and imbalanced healthcare datasets. Techniques such as Isolation Forest, Bayesian hierarchical models, and deep autoencoders demonstrate superior performance compared to traditional methods. Despite significant advancements, gaps remain with regard to transfer learning, interpretability and explainability of models, and the development of real-time, incremental learning algorithms. Future research should focus on these areas to enhance fraud detection accuracy and trust. Our work aims to provide a valuable resource for researchers and practitioners, supporting the development of more robust and adaptive fraud detection systems to protect healthcare integrity and reduce financial losses.

In closed-system domains, such as healthcare databases, record scarcity and data quality often act as barriers to applying state-of-the-art language processing techniques. Addressing these challenges requires the adjustment of both domain and task to effectively deliver meaningful value. A common approach for adapting domains with limited and poorly annotated data is data augmentation. Transformers and Sequential Denoising Auto-Encoders (TSDAEs) offer an inductive, unsupervised pretraining method that efficiently leverages unlabeled data by learning from many-to-one corrupted training samples. This approach reduces the need for extensive manual data annotation typically associated with domain adaptation. We advance this method by using transduction-based noise generation, which simulates the kind of noise commonly encountered in text generation within targeted domains. Our study investigates the effects of corruption and contextual noise introduced by this augmentation, thus enhancing the practical ability of domain-adapted models in specialized fields.

OCR2SEQ represents an innovative advancement in Optical Character Recognition (OCR) technology, leveraging a multi-modal generative augmentation strategy to overcome traditional limitations in OCR systems. This paper introduces OCR2SEQ’s unique approach, tailored to enhance data quality for sequence-to-sequence models, especially in scenarios characterized by sparse character sets and specialized vocabularies. At the heart of OCR2SEQ lies a set of novel augmentation techniques designed to simulate realistic text extraction errors. These techniques are adept at generating diverse and challenging data scenarios, thereby substantially improving the training efficacy and accuracy of text-to-text transformers. The application of OCR2SEQ has shown notable improvements in data processing accuracy, particularly in sectors heavily dependent on OCR technologies such as healthcare and library sciences. This paper demonstrates the capability of OCR2SEQ to transform OCR systems by enriching them with augmented, domain-specific data, paving the way for more sophisticated and reliable machine learning interpretations. This advancement in OCR technology, as presented in the study, not only enhances the accuracy and reliability of data processing but also sets a new benchmark in the integration of augmented data for refining OCR capabilities.

As machine learning continues to be a promising tool for cyber security, industry and researchers have continued to develop datasets for research. These datasets often contain multiple emulated exemplars for common attacks seen in real-world networks. The datasets provide researchers with the necessary samples to train and test the detection capabilities of their machine learning models. This paper contains an in-depth analysis of the composition of one of the newest datasets, Bot-IoT. The full dataset contains about 73 million instances (big data), three dependent features, 26 independent features, and four primary attack categories. The purpose of this paper is to provide researchers with an understanding of the environment used to create Bot-IoT and how that environment effected its composition. A detailed analysis of the dataset's composition can provide additional insight into the dataset's suitability for machine learning.

With the massive resources and strategies accessible to attackers, countering Denial of Service (DoS) attacks is getting increasingly difficult. One of these techniques is application-layer DoS. Due to these challenges, network security has become increasingly more challenging to ensure. Hypertext Transfer Protocol (HTTP), Domain Name Service (DNS), Simple Mail Transfer Protocol (SMTP), and other application protocols have had increased attacks over the past several years. It is common for application-layer attacks to concentrate on these protocols because attackers can exploit some weaknesses. Flood and “low and slow” attacks are examples of application-layer attacks. They target weaknesses in HTTP, the most extensively used application-layer protocol on the Internet. Our experiment proposes a generalized detection approach to identify features for application-layer DoS attacks that is not specific to a single slow DoS attack. We combine four application-layer DoS attack datasets: Slow Read, HTTP POST, Slowloris, and Apache Range Header. We perform a feature-scaling technique that applies a normalization filter to the combined dataset. We perform a feature extraction technique, Principal Component Analysis (PCA), on the combined dataset to reduce dimensionality. We examine ways to enhance machine learning techniques for detecting slow application-layer DoS attacks that employ these methodologies. The machine learners effectively identify multiple slow DoS attacks, according to our findings. The experiment shows that classifiers are good predictors when combined with our selected Netflow characteristics and feature selection techniques.

Training a machine learning algorithm on a class-imbalanced dataset can be a difficult task, a process that could prove even more challenging under conditions of high dimensionality. Feature extraction and data sampling are among the most popular preprocessing techniques. Feature extraction is used to derive a richer set of reduced dataset features, while data sampling is used to mitigate class imbalance. In this paper, we investigate these two preprocessing techniques, using a credit card fraud dataset and four ensemble classifiers (Random Forest, CatBoost, LightGBM, and XGBoost). Within the context of feature extraction, the Principal Component Analysis (PCA) and Convolutional Autoencoder (CAE) methods are evaluated. With regard to data sampling, the Random Undersampling (RUS), Synthetic Minority Oversampling Technique (SMOTE), and SMOTE Tomek methods are evaluated. The F1 score and Area Under the Receiver Operating Characteristic Curve (AUC) metrics serve as measures of classification performance. Our results show that the implementation of the RUS method followed by the CAE method leads to the best performance for credit card fraud detection.

Machine learning research on Medicare fraud detection is of national importance, primarily due to the extensive financial losses caused by this deceptive practice. Our big data study focuses on the Medicare Part D dataset, which we utilize to detect healthcare fraud perpetrated by physicians. In this paper, we compare and contrast One-Class Classification (OCC) and binary classification by examining eight different classifiers. The metrics applied in this analysis are Area Under the Receiver Operating Characteristic Curve (AUC) and Area Under the Precision-Recall Curve (AUPRC). Our findings indicate that binary classification outperforms OCC in Medicare fraud detection. Furthermore, we establish that the Decision Tree-based classifiers employed in the research are the most effective, with CatBoost delivering the best performance.

Research into machine learning methods for fraud detection is of paramount importance, largely due to the substantial financial implications associated with fraudulent activities. Our investigation is centered around the Credit Card Fraud Dataset and the Medicare Part D dataset, both of which are highly imbalanced. The Credit Card Fraud Detection Dataset is large data and contains actual transactional content, which makes it an ideal benchmark for credit card fraud detection. The Medicare Part D dataset is big data, providing researchers the opportunity to examine national trends and patterns related to prescription drug usage and expenditures. This paper presents a detailed comparison of One-Class Classification (OCC) and binary classification algorithms, utilizing eight distinct classifiers. OCC is a more appealing option, since collecting a second label for binary classification can be very expensive and not possible to obtain within a reasonable time frame. We evaluate our models based on two key metrics: the Area Under the Precision-Recall Curve (AUPRC)) and the Area Under the Receiver Operating Characteristic Curve (AUC). Our results show that binary classification consistently outperforms OCC in detecting fraud within both datasets. In addition, we found that CatBoost is the most performant among the classifiers tested. Moreover, we contribute novel results by being the first to publish a performance comparison of OCC and binary classification specifically for fraud detection in the Credit Card Fraud and Medicare Part D datasets.

The yearly increase in incidents of credit card fraud can be attributed to the rapid growth of e-commerce. To address this issue, effective fraud detection methods are essential. Our research focuses on the Credit Card Fraud Detection Dataset, which is a widely used dataset that contains real-world transaction data and is characterized by high class imbalance. This dataset has the potential to serve as a benchmark for credit card fraud detection. Our work evaluates the effectiveness of two supervised learning classification techniques, binary classification and one-class classification, for credit card fraud detection. The performance of five binary-class classification (BCC) learners and three one-class classification (OCC) learners is evaluated. The metrics used are area under the precision-recall curve (AUPRC) and area under the receiver operating characteristic curve (AUC). Our results indicate that binary classification is a better approach for detecting credit card fraud than one-class classification, with the top binary classifier being CatBoost.

Our study evaluates the impact of training OneClass Classification (OCC) algorithms on the majority class compared to training them on the minority class, using large and big datasets that are highly imbalanced. It is important to note that class availability can present a significant challenge in model training. In reality, there may be situations where one class is readily obtainable within a reasonable time frame while another class is not. Our task involves detecting instances of fraud in the Credit Fraud Detection Dataset and our Medicare dataset derived from Medicare Part D data and List of Excluded Individuals and Entities (LEIE) data. The Credit Card Fraud Detection Dataset has real-world transaction content as well as a significant class imbalance, making it suitable for use as a benchmark for credit card fraud detection. In addition, it is the only publicly available large data for credit card fraud analysis. Part D is big data, allowing researchers to analyze national trends and patterns in prescription drug usage and expenditures. The algorithms used in the study are One-Class Gaussian Mixture Model (GMM), OneClass Adversarial Nets (OCAN), and One-Class Support Vector Machine (SVM). Their performance is measured with the Area Under the Precision-Recall Curve (AUPRC) and Area Under the Receiver Operating Characteristic Curve (AUC). Our results indicate that OCC produces better results when models are trained on the majority class.

Output thresholding is well-suited for addressing class imbalance, since the technique does not increase dataset size, run the risk of discarding important instances, or modify an existing learner. Through the use of the Credit Card Fraud Detection Dataset, this study proposes a threshold optimization approach that factors in the constraint True Positive Rate (TPR) ≥ True Negative Rate (TNR). Our findings indicate that an increase of the Area Under the Precision–Recall Curve (AUPRC) score is associated with an improvement in threshold-based classification scores, while an increase of positive class prior probability causes optimal thresholds to increase. In addition, we discovered that best overall results for the selection of an optimal threshold are obtained without the use of Random Undersampling (RUS). Furthermore, with the exception of AUPRC, we established that the default threshold yields good performance scores at a balanced class ratio. Our evaluation of four threshold optimization techniques, eight threshold-dependent metrics, and two threshold-agnostic metrics defines the uniqueness of this research.

Practitioners and researchers of machine learning should have a deep understanding about the selection of the right performance metrics for classifier evaluation. Using a credit card fraud dataset, we demonstrate that the Area Under the Precision-Recall Curve (AUPRC) metric is a more reliable measurement, for the classification of highly imbalanced data, than the Area Under the Receiver Operating Characteristic Curve (AUC) metric. Furthermore, we establish that AUC is minimally impacted by the use of Random Undersampling (RUS). The classifiers used in this study are ensemble learners: LightGBM, CatBoost, Extremely Randomized Trees (ET), XGBoost, and Random Forest. Our results are governed by the fact that in a highly imbalanced dataset, the comparatively large number of true negative instances has an influence on AUC but not on AUPRC. Hence, AUPRC is able to accurately detect changes in the number of false positives because it ignores the true negatives.

It is inherently challenging to train a machine learning algorithm on a class-imbalanced dataset. Under conditions of high dimensionality, this training process can become even more difficult due to the large number of features in the dataset. During preprocessing, data sampling is commonly used to address class imbalance and feature extraction is frequently used to reduce the number of dataset features. In this study, we explore the use of these two preprocessing activities before passing on the data to four ensemble classifiers (Random Forest, CatBoost, LightGBM, and XGBoost). With reference to feature extraction, the Principal Component Analysis (PCA) and Convolutional Autoencoder (CAE) methods are evaluated. With regard to data sampling, the Random Undersampling (RUS) and Synthetic Minority Oversampling Technique (SMOTE) methods are evaluated. Classification performance is measured with the Area Under the Receiver Operating Characteristic Curve (AUC) metric. Our results indicate that the implementation of the RUS method followed by the CAE method leads to the best classification performance.

Bot-IoT is a recent and publicly available dataset that depicts attack traffic launched by BotNets against internet of things (IoT) networks. Normal (non-attack) traffic is represented by over 9,000 of the approximately 73,000,000 instances of big data that constitute this dataset. We present an easy-to-learn Bot-IoT approach, centred on the use of a minimum number of dataset features and a simple machine learning algorithm. Our contribution is defined by decision tree models built from derived Bot-IoT datasets with no more than three features. As per our definition of easy-to-learn, we require that predictive models have area under the receiver operating characteristic curve (AUC) mean scores greater than 0.99. According to our results, all the derived datasets produce easy-to-learn models. To the best of our knowledge, this work, in terms of its simplicity, interpretability, and performance, is an improvement over Bot-IoT classification approaches in existing literature.