Annals of Data Science最新文献

{"title":"Investigating the Performance of the Super-Learning Ensemble Algorithm in the Binary Classification of Diseases","authors":"Toluwase Ayobami Olowookere,&nbsp;Aanuoluwa Oyebola Adio,&nbsp;Godbless Uwen Peter,&nbsp;Olufunmilayo I. Olopade,&nbsp;Benjamin S. Aribisala","doi":"10.1007/s40745-026-00686-4","DOIUrl":"10.1007/s40745-026-00686-4","url":null,"abstract":"<div><p>The super-learning ensemble algorithm has been demonstrated in the literature to improve classification tasks. However, limited works have been presented in the literature on investigating the performance of the super-learning ensemble algorithm on various benchmark datasets of diseases. This study, therefore, seeks to investigate and establish the effectiveness of the super learning algorithm in the binary classification of diseases. Five binary-class benchmark datasets of diseases from Kaggle and UCI repositories were considered in this study. Results show that the super-learning algorithm produces a consistently improved classification model over the base learning ensemble algorithms or as good as the base learning ensemble algorithms. On the breast cancer dataset, the super-learning algorithm achieved 97.7% accuracy and sensitivity of 0.968, which is the same performance as the best ensemble base-learner (AdaBoost). Also, on the diabetes dataset, the super-learning algorithm achieved 80.7% accuracy and a sensitivity of 0.834, while the best base ensemble learner (Bagging) has an accuracy of 80.3% and a sensitivity of 0.781. For heart failure data, the super-learner model algorithm achieved 87.7% accuracy and a sensitivity of 0.859 while the best base ensemble learner (AdaBoost) has 86.6% accuracy and 0.841 sensitivity. In the prostate cancer dataset, the super-learner algorithm achieved 54.7% accuracy and a sensitivity of 0.519 while the best base ensemble learner (Random Forest) achieved 53.0% accuracy and 0.410 sensitivity. Lastly, on the stroke dataset, the super-learning algorithm, however, achieved 96.7% accuracy and a sensitivity of 0.950 while the best base ensemble learner (Random Forest) has an accuracy of 96.8% and a sensitivity of 0.944. The performance of the work presented confirms that the super-learner algorithm has the ability to better the classification performance of the base learning ensemble algorithms or perform at the same level of accuracy with them for binary classification of diseases tasks.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 3","pages":"609 - 625"},"PeriodicalIF":0.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148458670","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparative Analysis of Machine Learning and Deep Learning Models of Binary Classification of Alzheimer Disease Using Longitudinal Non-imaging Data","authors":"Bukola Badeji-Ajisafe,&nbsp;Balikis Pelumioluwa Abdulraheem","doi":"10.1007/s40745-026-00679-3","DOIUrl":"10.1007/s40745-026-00679-3","url":null,"abstract":"<div><p>The brain is a mass of nerve tissue that controls all of our bodily and cognitive functions. It is susceptible to neurodegenerative diseases such as Alzheimer’s disease (AD). AD is a progressive and irreversible disease that leads to memory loss, continuous cognitive decline, and behavioral deterioration. Early identification of AD stages is crucial for timely intervention and improved patient care. This research compares the performance of multiple Artificial Intelligence (AI) algorithms trained with a longitudinal non-imaging dataset to classify individuals as either ‘Nondemented’ or ‘Demented’. The Support Vector Model (SVM), which is termed long_SVC, achieved the highest performance with an accuracy of 0.829 and an Area under Curve (AUC) of 0.830. The Random Forest model (long_RFC) was the second best, achieving an accuracy of 0.808 and an AUC score of 0.806. Deep learning (DL) approaches, including the long_1D-CNN model and long_LSTM-AE model, performed poorly in differentiating between ‘Non-demented’ and ‘Demented’. The long_1D-CNN model performed better than the long_LSTM-AE model; it had an accuracy of 0.566 and an AUC score of 0.585. The long_LSTM-AE has an accuracy of 0.533 and an AUC score of 0.558. The long_1D-CNN and long_LSTM-AE models had a log loss of 0.683 and 0.708, respectivel. This demonstrated the limited effectiveness of DL algorithms on small datasets. The outcomes showed that traditional machine learning approaches can outperform DL methods when working with limited longitudinal data. This research highlights the potential of the long_SVC model as a decision support tool in healthcare, helping doctors to come to an early diagnosis. Challenges related to real-world deployment, such as model interpretability for doctors, are discussed. Future studies could explore integrating this model into real-world healthcare settings to improve patient outcomes.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 3","pages":"575 - 590"},"PeriodicalIF":0.0,"publicationDate":"2026-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148458672","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Roles of Big Data Analytics on Customer Experience in eCommerce and Digital Financial Platforms","authors":"Kolawole G. Akintola,&nbsp;Bukola A. Ifedayo-Ojo,&nbsp;Olukemi V. Olatunde,&nbsp;Oluwadamilola E. Muritala,&nbsp;Saidi O. Lawal","doi":"10.1007/s40745-025-00672-2","DOIUrl":"10.1007/s40745-025-00672-2","url":null,"abstract":"<div><p>The rapid generation of structured and unstructured data—commonly referred to as big data—offers great opportunities for financial institutions and systems, such as electronic banking (e-banking), electronic commerce (e-commerce) and electronic Payment systems, to extract valuable insights that enhance operational efficiency, profitability and enhanced customer experience. These data are obtained from customers, sellers, shopping, and business activities. In this work, a review of big data analytics techniques applied to e-commerce and digital financial services (DFS) in this era of digital economy is carried out. The review attempts to demonstrate the role of big data analytics in financial data analysis and ecommerce in the contemporary digital economy. The research also examines the role of big data analytics in enhancing customer experience in ecommerce. Three hypotheses were formulated as: effect of big data analytics on predicting users’ experience on different platforms leading to customer satisfaction, effect of big data analytics on predicting users’ experience based on operational efficiency on different platforms. Effect of big data analytics in predicting customers retention based on operational efficiency on different platforms. The results obtained showed that big data analytics can actually have impact on users’ experience, but it may not be the only reason as there some other factors that determine users experience in ecommerce platforms.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 3","pages":"727 - 739"},"PeriodicalIF":0.0,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148458945","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Process Model Development for Recovery of Heavy Metals from Wastewater Using Artificial Intelligence","authors":"Abiola Ezekiel Taiwo,&nbsp;Paul Musonge","doi":"10.1007/s40745-026-00681-9","DOIUrl":"10.1007/s40745-026-00681-9","url":null,"abstract":"<div><p>Industrial and economic growth increasingly lead to the generation of wastewater laden with diverse pollutants, posing significant threats to natural water resources. Artificial intelligence (AI) offers powerful capabilities for data analysis, classification, and prediction, particularly excelling in solving the complex, nonlinear problems prevalent in water-related research compared to conventional modeling approaches. Due to their extreme toxicity, the removal of heavy metals from water and wastewater has garnered significant research focus. Consequently, AI offers a promising avenue for developing sustainable and efficient technologies for water and wastewater treatment. As part of the wastewater treatment process, this study demonstrates how artificial intelligence modelling techniques can predict the amount of adsorbed heavy metals. Results from the case study scenario show approximately 95–98% maximum removal efficiency of zinc and lead in the wastewater. These findings confirm that Artificial Neural Networks (ANN) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are robust and reliable tools for modeling heavy metal removal from wastewater. In addition, AI will assist stakeholders in making decisions regarding wastewater treatment.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 3","pages":"649 - 663"},"PeriodicalIF":0.0,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148458856","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Automatic Misogyny Detection using Supervised Artificial Neural Networks","authors":"Olanike Christianah Akinduyite,&nbsp;Ogheneruno Chris-Umoru","doi":"10.1007/s40745-025-00673-1","DOIUrl":"10.1007/s40745-025-00673-1","url":null,"abstract":"<div><p>Modern developments in social networks and internet technologies have drawn researchers from around the world to investigate cyber bullying meme detection systems for detecting targeted bullying text and images in females, known as misogynistic text and images. Based on the reviewed literature, very few studies have employed BiLSTM and VVCG16 models on the multimedia automatic misogynistic identification (MAMI) dataset via bidirectional long-short-term memory (BiLSTM) and visual geometry group 16 (VGG16). Hence, this study aims to develop a misogynistic text and image detection system using bidirectional long short-term memory (BiLSTM) for text classification and visual geometry group 16 (VGG16) for image classification. The methodology integrates the visual geometry group 16 (VGG16) and bidirectional long-short-term memory (BiLSTM) models for multimedia analysis. The MAMI dataset, which includes 12,000 misogynistic memes split into training and test sets in the ratio of 75% to 25%, is utilized. Furthermore, this study addresses two primary subtasks. Subtask A involves classifying memes either as misogynistic or nonmisogynistic, and subtask B aims at identifying the type of misogyny, which includes stereotypes, objectification, shaming, and violence. The analysis and visualization of these categories were conducted via pandas, NumPy, spacy, unicodedata and scikit learning. The performance of the system was enhanced through the application of both the CNN and RNN algorithms on the MAMI dataset, which yielded the best performance, with an accuracy of 0.99 on both text and image data.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 3","pages":"681 - 700"},"PeriodicalIF":0.0,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148458879","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An Integrated Pythagorean Fuzzy Delphi-AHP Framework for Optimizing Foreign Direct Investment: Key Drivers for Success","authors":"Srikant Gupta,&nbsp;Babita Jha,&nbsp;Gaurav Dawar","doi":"10.1007/s40745-025-00667-z","DOIUrl":"10.1007/s40745-025-00667-z","url":null,"abstract":"<div><p>Foreign Direct Investment (FDI) plays a pivotal role in global economic development, fostering cross-border collaborations and driving economic growth. Recognizing the significance of optimizing FDI drivers, this study employs a novel approach by integrating the Pythagorean Fuzzy Delphi (PFD) and Pythagorean Fuzzy Analytic Hierarchy Process (PFAHP). The Pythagorean Fuzzy Delphi methodology was used to identify and classify drivers into Technological, Political, Environmental, Social, and Cultural categories. Subsequently, the PFAHP was employed to rank these drivers. The top three prioritized drivers are: advocating for favorable foreign investment policies and trade agreements; implementing advanced cybersecurity measures to safeguard sensitive technology and data; and developing cutting-edge research and development facilities to foster innovation and attract technology-intensive investments. The study concludes by discussing how implementing these top-ranked drivers can significantly enhance FDI by creating a conducive environment for international investment, thereby contributing to economic prosperity and technological advancement.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 6","pages":"1605 - 1641"},"PeriodicalIF":0.0,"publicationDate":"2025-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149230744","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cancelable Speaker Identification Based on Speech Deconvolution Methods","authors":"Marwa A. Elsayed,&nbsp;Walid El-Shafai,&nbsp;Mohsen A. Rashwan,&nbsp;Moawad I. Dessouky,&nbsp;Adel S. El-Fishawy,&nbsp;Fathi E. Abd El-Samie","doi":"10.1007/s40745-024-00579-4","DOIUrl":"10.1007/s40745-024-00579-4","url":null,"abstract":"<div><p>Biometric authentication systems, which use unique biological traits for identification, have gained popularity in various fields as a replacement for traditional password- or token-based systems. While offering enhanced security, these systems remain vulnerable to hacking attempts. To address this concern, this paper presents a novel masking technique for speaker identification, enhancing security and privacy. The core concept involves generating an alternate version of the original speaker template using a one-way transformation. This ensures that the original template remains protected and unused within the system. The transformation draws inspiration from deconvolution methods, commonly used for signal processing. These methods are applied in the presence of noise to magnify the noise in order to mask the original templates. We explore three specific deconvolution methods for creating cancelable biometric templates, namely Linear Minimum Mean Square Error (LMMSE), inverse filter and regularized deconvolution. The LMMSE method depends on statistical techniques to minimize the error between the original signal and the deconvolution output. The paper discusses some assumptions that help reduce the computational complexity of the LMMSE solution. Inverse filter deconvolution directly reverses the convolution process through multiplication with the inverse filter transfer function in frequency domain. Finally, regularized deconvolution incorporates additional constraints during the deconvolution process to improve the robustness of the transformed templates. Additionally, we investigate two further methods for comparison: Discrete Wavelet Transform (DWT) and wavelet thresholding. These methods offer alternative approaches to signal processing and feature extraction. Simulation results demonstrate the effectiveness of the deconvolution-based cancelable biometric schemes, displaying their potential for enhancing the security and privacy of speaker identification systems. A novel finding from this research is that the inverse filter deconvolution method, despite its known limitations, offers superior performance in comparison to other methods when applied in cancelable speaker identification. This unexpected outcomes change the conventional understanding of the method applicability and opens new directions for research in the field.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 1","pages":"1 - 27"},"PeriodicalIF":0.0,"publicationDate":"2025-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147342382","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Revisiting Computing Paradigm for Partitional Clustering Analysis and Recommendations","authors":"Hakam Singh,&nbsp;Yugal Kumar,&nbsp;Nagesh Kumar","doi":"10.1007/s40745-025-00663-3","DOIUrl":"10.1007/s40745-025-00663-3","url":null,"abstract":"<div><p>The business world is concentric around the imperative of information extraction and analysis. The information extraction processes are blended with different data mining techniques like clustering, classification, etc. Clustering is an explorative technique that extracts imperative information from large databases. The adaptability of clustering methods in various disciplines leads to momentous research; immersive work is in this field. Several algorithms are reported for clustering to handle dissimilar clustering problems like initialization, local optima, diversity, and convergence rate. In addition to this, clustering methods are automated, improved/hybridized to obtain new/robust clustering algorithms. An extensive literature survey was carried out to provide in-depth knowledge of partitional clustering algorithms, datasets, and performance measures. This study considers the last twelve years substantial work on partitional clustering and well summarized to promote profound understanding and in-depth learning. The outcome of this research would highlight the key aspects determining the most suitable partitioning and clustering approaches, in turn influencing future work. Furthermore, it offers valuable insights for researchers working in the domain of clustering and data analytics.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 2","pages":"423 - 453"},"PeriodicalIF":0.0,"publicationDate":"2025-11-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147607328","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive Hierarchical Attention for Multivariate Time Series Anomaly Detection","authors":"Xiaohan You,&nbsp;Xiaobo Guo,&nbsp;Binfeng Wang,&nbsp;Changbin Wang","doi":"10.1007/s40745-025-00662-4","DOIUrl":"10.1007/s40745-025-00662-4","url":null,"abstract":"<div><p>In multivariate time series (MTS) anomaly detection, existing graph neural network (GNN) methods often neglect multi-level feature representations, relying solely on the final layer output and fixed thresholds, which leads to information loss and high false-positive rates. To address this issue, we propose a MTS adaptive hierarchical attention (MTS-AHA) model that integrates cross-layer features from local to global topological structures across GNN layers. Additionally, we introduce a dynamic threshold module that adaptively adjusts the threshold by fitting the tail features of the distribution. Extensive experiments on three datasets validate the superiority of our method and its components.</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"13 2","pages":"401 - 421"},"PeriodicalIF":0.0,"publicationDate":"2025-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147606498","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"System of Systems Engineering Foundation","authors":"Mo Jamshidi","doi":"10.1007/s40745-025-00652-6","DOIUrl":"10.1007/s40745-025-00652-6","url":null,"abstract":"<div><p>This paper introduces the concept of system of systems (SoS) and the challenges ahead to extend systems engineering (SE) to system of systems engineering. The birth of a new engineering field may be on the horizon - System of Systems Engineering (SoSE). A SoS is a collection of individual, possibly heterogeneous, but functional systems integrated together to enhance the overall robustness, lower the cost of operation, and increase reliability of the overall complex (SoS) system. Having said that the field has large vacuum from basic definition, to theory, to management and implementation. Many key issues like architecture, modeling, simulation, identification, emergence, standards, net-centricity, control, etc. are all begging for attention. In this review paper, we will be going through all these issues briefly and bring out the challenges to the attention of interested readers. This paper is based on a keynote presented at the ITQM International Conference, Bucharest, Romania, August 23-24, 2024</p></div>","PeriodicalId":36280,"journal":{"name":"Annals of Data Science","volume":"12 6","pages":"1857 - 1881"},"PeriodicalIF":0.0,"publicationDate":"2025-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145537717","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}