Machine learning (ML) algorithms have gained prominence in time series prediction problems. Depending on the nature of the time series data, it can be difficult to build an accurate ML model with the proper structure and hyperparameters. In this study, we propose a predictive error compensation wavelet neural network model (PEC-WNN) for improving the prediction accuracy of chaotic and stochastic time series data. In the proposed model, an additional network is used for the prediction of the main network error to compensate the overall prediction error. The main network takes as inputs the time series data through moving frames in multiple-scales. The same structure and hyperparameter sets are applied for quite distinct four types of problems for verification of the robustness and accuracy of the proposed model. Specifically, the Mackey-Glass, Box-Jenkins, and Lorenz Attractor benchmark problems, as well as drought forecasting are used to characterize the performance of the model for chaotic and stochastic data cases. The results show that the PEC-WNN provides significantly more accurate predictions for all compared benchmark problems with respect to conventional machine learning and time series prediction methods without changing any hyperparameter or the structure. In addition, the time and space complexity of the PEC-WNN model is less than all other compared ML methods, including long short-term memory (LSTM) and convolutional neural networks (CNNs).

Alzheimer's disease represents the most common age-related neurodegenerative disorder and a leading cause of progressive cognitive impairment. Predicting cognitive decline is challenging but would be invaluable in an increasingly aging population which also experiences a rising cardiovascular risk. In order to examine whether plasma measurements of one of the established biomarkers of heart failure, brain natriuretic peptide (BNP), reflect a decline in cognitive function, associated with Alzheimer's disease neurodegeneration, BNP levels were analysed, by using a novel assay called a SOMAscan, in 1. cognitively healthy, control subjects; 2. subjects with mild cognitive impairment, and 3. subjects with Alzheimer's disease. The results of our study show that the levels of the BNP were significantly different between the three types of diagnoses (p < 0.05), whereby subjects with mild cognitive impairment had the lowest mean BNP value, and healthy subjects had the highest BNP value. Importantly, our results show that the levels of the BNP are influenced by the presence of at least one APOE4 allele in the healthy (p < 0.05) and in the Alzheimer's disease groups of subjects (p < 0.1). As the levels of the BNP appear to be independent of the APOE4 genotype in subjects with mild cognitive impairment, the results of our study support inclusion of measurements of plasma levels of the BNP in the list of the core Alzheimer's disease biomarkers for identification of the mild cognitive impairment group of patients. In addition, the results of our study warrant further investigations into molecular links between Alzheimer's disease-type cognitive decline and cardiovascular disorders.

Multivariable machine learning (ML) models are increasingly used for time series predictions. However, avoiding the overfitting and underfitting in ML-based time series prediction requires special consideration depending on the size and characteristics of the available training dataset. Predictive error compensating wavelet neural network (PEC-WNN) improves the time series prediction accuracy by enhancing the orthogonal features within a data fusion scheme. In this study, time series prediction performance of the PEC-WNNs have been evaluated on two different problems in comparison to conventional machine learning methods including the long short-term memory (LSTM) network. The results have shown that PECNET provides significantly more accurate predictions. RMSPE error is reduced by more than 60% with respect to other compared ML methods for Lorenz Attractor and wind speed prediction problems.

The rapid spread of fake news in the digital age poses significant challenges, necessitating effective detection methods. This study presents a comprehensive evaluation of various ensemble and machine learning classifiers, combined with different feature selection techniques, to improve the accuracy and reliability of the detection of fake news. Using the TruthSeeker dataset, this research examines feature selection methods such as Recursive Feature Elimination (RFE), SelectKBest, Principal Component Analysis (PCA) and Genetic Algorithms (GA), analyzing their impact on model performance. Key metrics such as accuracy, precision, recall, F1 score, and AUC-ROC were used to assess the effectiveness of each classifier. The results reveal that ensemble methods, particularly Random Forest (RF) and Gradient Boosting, demonstrate superior performance, achieving high accuracy and AUC-ROC scores. Moreover, feature selection techniques like RFE and SelectKBest significantly improve model outcomes by optimizing the feature set, while PCA is less effective in this context. This study highlights the importance of integrating robust classifiers with optimal feature selection methods to improve the efficacy of fake news detection systems.

Phishing websites remain a persistent cybersecurity threat because they can be rapidly generated, visually disguised, and adapted to evade static detection mechanisms. Although machine learning-based phishing detection has shown strong performance, many existing approaches rely on offline training or fixed model selection, which may be unsuitable for continuously evolving streams. Incremental learning addresses this limitation by allowing models to update continuously as new instances arrive. However, selecting a single incremental learner before deployment may be restrictive, while using a full ensemble can increase prediction latency, update time, and memory demand. To address this challenge, this paper investigates a multi-armed bandit (MAB)-based online model selection framework for incremental phishing detection. In the proposed framework, heterogeneous incremental learning pipelines are treated as candidate arms, and a bandit policy dynamically selects one learner for each incoming instance based on observed rewards. Several MAB policies are evaluated, including Random selection, Epsilon-Greedy, UCB, Bayes-UCB, Thompson Sampling, EXP3, and LinUCB. The study further compares three update strategies: selected-only update, hybrid update, and all-model update. The framework is evaluated against individual incremental learners and a full ensemble baseline using accuracy, precision, recall, F1-score, Cohen’s kappa, timing decomposition, memory footprint, and arm selection behavior. The results show that MAB-based selection provides a flexible trade-off between predictive reliability and computational efficiency. The strongest standalone learner remains highly competitive, but hybrid and all-model MAB configurations approach its performance while offering adaptive model selection. Selected-only MAB provides the lowest processing cost, all-model MAB provides stronger predictive reliability at higher computational cost, and hybrid MAB offers the most practical balance. The findings highlight the value of MAB-based online model selection as a resource-aware mechanism for adaptive phishing detection in streaming environments.