Data Transformation Research Articles

AI-Driven Data Preparation: The Key to Unlocking Cloud-Based Analytics

The rapid adoption of cloud-based analytics has revolutionized data-driven decision-making across industries. Cloud- based analytics has transformed how businesses make decisions by leveraging vast amounts of data. However, preparing data for analysis—such as cleaning, transforming, and organizing it—can be a complicated and time- consuming process. AI-driven data preparation (AIDP) is a solution that automates these steps, reducing the time and effort needed to prepare data while improving its quality. This paper explains the importance of AI-driven data preparation, discusses how it works, and shows how businesses can benefit from using AI in their data preparation process for cloud analytics. The use of diagrams, flowcharts, and pseudocode helps explain these concepts in a simplified yet technical manner. Keywords AI-driven data preparation, cloud-based analytics, data pipeline, automation, machine learning, data wrangling, ETL, data transformation, big data

HERALD: A domain-specific query language for longitudinal health data analytics

BackgroundLarge-scale health data has significant potential for research and innovation, especially with longitudinal data offering insights into prevention, disease progression, and treatment effects. Yet, analyzing this data type is complex, as data points are repeatedly documented along the timeline. As a consequence, extracting cross-sectional tabular data suitable for statistical analysis and machine learning can be challenging for medical researchers and data scientists alike, with existing tools lacking balance between ease-of-use and comprehensiveness. ObjectiveThis paper introduces HERALD, a novel domain-specific query language designed to support the transformation of longitudinal health data into cross-sectional tables. We describe the basic concepts, the query syntax, a graphical user interface for constructing and executing HERALD queries, as well as an integration into Informatics for Integrating Biology and the Bedside (i2b2). MethodsThe syntax of HERALD mimics natural language and supports different query types for selection, aggregation, analysis of relationships, and searching for data points based on filter expressions and temporal constraints. Using a hierarchical concept model, queries are executed individually for the data of each patient, while constructing tabular output. HERALD is closed, meaning that queries process data points and generate data points. Queries can refer to data points that have been produced by previous queries, providing a simple, but powerful nesting mechanism. ResultsThe open-source implementation consists of a HERALD query parser, an execution engine, as well as a web-based user interface for query construction and statistical analysis. The implementation can be deployed as a standalone component and integrated into self-service data analytics environments like i2b2 as a plugin. HERALD can be valuable tool for data scientists and machine learning experts, as it simplifies the process of transforming longitudinal health data into tables and data matrices. ConclusionThe construction of cross-sectional tables from longitudinal data can be supported through dedicated query languages that strike a reasonable balance between language complexity and transformation capabilities.

Real Time Data Ingestion and Transformation in Azure Data Platforms

Real Time Data Ingestion and Transformation in Azure Data Platforms

Positioning of the test surface in a CGH null test by cat's eye interference

Interferometric null test of optical aspheres and freeforms is one of the most established methods, in spite of recognized limitations related to positioning of the test surface. Slight misalignment of the surface in the null test system can introduce remarkable wave aberrations. Current approaches for positioning the test surface are based on a series of fiducial marks or retroreflectors set at given spots around the test surface, which consequently suffer the problem of datum transformation. We propose a method for positioning the test surface in a null test by cat's eye interference without any fiducial marks or retroreflectors. A computer-generated hologram (CGH) is used as null optics fabricated with null test pattern, alignment pattern and positioning pattern. A part of test beam with relatively small f/number is diffracted through the positioning pattern and then focuses on certain spots of the test surface. The cat's eye reflection from the test surface returns to the interferometer and interferes with the reference beam. It is then possible to precisely position the test surface by measuring the wavefront error of the cat's eye interference. The design method for such a CGH positioning pattern is presented, following the physical constraint that the surface normal at the cay's eye reflection is right the angular bisector of the positioning beam. Analysis on the positioning performance is then presented to show the sensitivity to misalignment including defocus, lateral shift and tip-tilt, which at last is experimentally verified by measuring an even asphere with a CGH.

Optimal designs for linear regression models with skew-normal errors

This paper investigates optimal designs for linear regression models with skew-normal errors which provide an alternative to traditional models with symmetric normal errors and eliminate the need for data transformation. Using the maximum likelihood estimation approach, we consider various design criteria, including D-, A-, A s -optimality. Equivalence theorems are provided to determine the optimality of the designs. Optimal designs for a centered parametrization model are also considered. Two examples are provided to illustrate the theoretical results. In addition, the efficiencies of the optimal designs based on skew-normal errors are compared with those based on normal errors.

Optimizing a Dynamic Vehicle Routing Problem with Deep Reinforcement Learning: Analyzing State-Space Components

Background: The dynamic vehicle routing problem (DVRP) is a complex optimization problem that is crucial for applications such as last-mile delivery. Our goal is to develop an application that can make real-time decisions to maximize total performance while adapting to the dynamic nature of incoming orders. We formulate the DVRP as a vehicle routing problem where new customer requests arrive dynamically, requiring immediate acceptance or rejection decisions. Methods: This study leverages reinforcement learning (RL), a machine learning paradigm that operates via feedback-driven decisions, to tackle the DVRP. We present a detailed RL formulation and systematically investigate the impacts of various state-space components on algorithm performance. Our approach involves incrementally modifying the state space, including analyzing the impacts of individual components, applying data transformation methods, and incorporating derived features. Results: Our findings demonstrate that a carefully designed state space in the formulation of the DVRP significantly improves RL performance. Notably, incorporating derived features and selectively applying feature transformation enhanced the model’s decision-making capabilities. The combination of all enhancements led to a statistically significant improvement in the results compared with the basic state formulation. Conclusions: This research provides insights into RL modeling for DVRPs, highlighting the importance of state-space design. The proposed approach offers a flexible framework that is applicable to various variants of the DVRP, with potential for validation using real-world data.

Mamba-in-Mamba: Centralized Mamba-Cross-Scan in Tokenized Mamba Model for Hyperspectral image classification

Hyperspectral image (HSI) classification plays a crucial role in remote sensing (RS) applications, enabling the precise identification of materials and land cover based on spectral information. This supports tasks such as agricultural management and urban planning. While sequential neural models like Recurrent Neural Networks (RNNs) and Transformers have been adapted for this task, they present limitations: RNNs struggle with feature aggregation and are sensitive to noise from interfering pixels, whereas Transformers require extensive computational resources and tend to underperform when HSI datasets contain limited or unbalanced training samples. To address these challenges, Mamba architectures have emerged, offering a balance between RNNs and Transformers by leveraging lightweight, parallel scanning capabilities. Although models like Vision Mamba (ViM) and Visual Mamba (VMamba) have demonstrated improvements in visual tasks, their application to HSI classification remains underexplored, particularly in handling land-cover semantic tokens and multi-scale feature aggregation for patch-wise classifiers. In response, this study introduces the Mamba-in-Mamba (MiM) architecture for HSI classification, marking a pioneering effort in this domain. The MiM model features: (1) a novel centralized Mamba-Cross-Scan (MCS) mechanism for efficient image-to-sequence data transformation; (2) a Tokenized Mamba (T-Mamba) encoder that incorporates a Gaussian Decay Mask (GDM), Semantic Token Learner (STL), and Semantic Token Fuser (STF) for enhanced feature generation; and (3) a Weighted MCS Fusion (WMF) module with a Multi-Scale Loss Design for improved training efficiency. Experimental results on four public HSI datasets—Indian Pines, Pavia University, Houston2013, and WHU-Hi-Honghu—demonstrate that our method achieves an overall accuracy improvement of up to 3.3%, 2.7%, 1.5%, and 2.3% over state-of-the-art approaches (i.e., SSFTT, MAEST, etc.) under both fixed and disjoint training-testing settings.

Hybridization of Machine Learning Model with Bee Colony based Feature Selection for Medical Data Classification

Nowadays, an important count of biomedical data is created continuously in several biomedical equipment and experiments because of quick technical enhancements in biomedical science. The study of clinical and health data is vital to enhance the analysis precision, prevention, and treatment. Initial analysis and treatment are extremely important approaches for preventing deaths in many diseases. Accordingly, the data mining and machine learning (ML) approaches are helpful tools for utilizing minimization error and for providing helpful data for analysis. But the data obtained in digital machines takes higher dimensionality, and not every data attained in digital machines is significant to specific diseases. This article develops an artificial bee colony-based feature selection with optimal hybrid ML model for medical data classification (ABCFS-OHML) technique. The presented ABCFS-OHML technique mainly aims to identify and classify the presence of disease using medical data. To attain this, the presented ABCFS-OHML technique initially pre-processes the input data in two ways namely null value removal and data transformation. Furthermore, the presented ABCFS-OHML technique uses ABCFS model for the choice of effectual subset of features. At last, root means square propagation with convolutional neural network-Hop field neural network (CNN-HFNN) model for classification purposes. The usage of RMSProp optimizer assists in attaining optimal hyperparameter selection of the CNN-HFNN method. The performance validation of the ABCFS-OHML technique takes place using three medical datasets. The comparison study reported that the ABCFS-OHML technique has accurately classified the medical data over other recent approaches.

Assessment of water quality and emerging pollutants in two fish species from the mallorquin swamp in the Colombian Caribbean

The Mallorquín Swamp, an important ecosystem in Atlántico, Colombian Caribbean, underwent environmental monitoring at eight points during rainy, transition, and dry seasons. This was to assess water quality, seasonal variation, and the bioaccumulation of metals, emerging pollutants, and organic compounds in the fish Ariopsis canteri and Mugil incilis. Water parameters were analyzed using descriptive statistics and multifactorial ANOVA with the Tukey HSD test for seasonal differences. Normality and variance of the fish results were verified, and differences between groups were evaluated using ANOVA or Kruskal-Walli's method when data transformation failed. Spearman correlation was used to relate the results. Water sampling revealed variations in temperature, dissolved oxygen, salinity, and nutrient levels. Significant differences in alkalinity and hardness were observed across seasons and sample points. The most probable number (MPN) levels of Total coliform and E. coli peaked near areas with domestic wastewater inputs, reaching 5.4x106 and 4.0x106 MPN, respectively, indicating potential microbiological contamination of water. Fish samples revealed high concentrations of persistent substances such as methylmercury, polycyclic aromatic hydrocarbons (PAHs), and emerging pollutants. Heavy metal analysis showed elevated iron levels (5.28 ± 0.657 mg/L), while emerging pollutants, including ibuprofen (218 μg/L) and naproxen (343.89 μg/L), exhibited high concentrations near human settlements. Ariopsis canteri showed higher bioconcentration tendencies for methylmercury (238.5 ± 100 μg/kg), and acenaphthene (7782 ± 4123.8 μg/kg), possibly influenced by its feeding habits and habitat preferences. In contrast, Mugil incilis exhibited higher bioaccumulation trends of PAH (2376.23 ± 599.63 μg/kg acenaphthene) and emerging pollutants like galaxolide (139.49 ± 34.98 μg/kg), possibly due to its mobility and exposure to various contaminants in their environment. These findings emphasize the need to monitor and manage aquatic ecosystems' health to mitigate anthropogenic influences on water quality and biodiversity. This research serves as a reference for global conservation efforts, emphasizing the need for comprehensive monitoring and regulatory frameworks to protect aquatic environments and ensure their sustainability for future generations.

Adaptive compressed learning boosts both efficiency and utility of differentially private federated learning

In the federated learning (FL) research field, current research is confronted with several pivotal challenges, e.g., data privacy, model utility and communication efficiency. Furthermore, these challenges are further amplified by statistical data heterogeneous in the FL system. Thus, a novel Communication-efficient and Utility-assured Gaussian differential privacy-based Personalized Federated Adaptive Compressed Learning method, called CUG-PFACL, is proposed. Specifically, an end-to-end local adaptive compressed learning strategy is designed, including three crucial modules, namely the measurement matrix, the personalized compressed data transformation and the local model. Especially, jointly training the measurement matrix module and the personalized compressed data transformation module can mitigate the inherent statistical heterogeneity while preserving all important characteristics of the compressed private data of each local client, and alleviate the additional heterogeneity induced by Gaussian differential privacy in each global communication round. Numerous experimental simulation and comparisons demonstrate that CUG-PFACL has three notable advantages: data privacy guarantee, enhanced personalized model utility and high-efficient communication.

High-resolution time-series classification in smart manufacturing systems

High-resolution time-series data is increasingly prevalent in today’s smart manufacturing systems driven by the proliferation of sensors and data acquisition technologies. However, classification tasks on such data face challenges like collinearity, model instability, poor generalization, and overfitting. This paper investigates techniques capable of handling high-resolution time-series classification (TSC) tasks in the manufacturing domain. High-resolution time-series data presents challenges impacting the concepts essential for classification algorithms. Additionally, increased noise and the extensive feature space further complicate classification performance, emphasizing the need for innovative solutions to process this data efficiently on resource-constrained devices. In this paper, we address this research gap by defining the problem in depth, investigating possible solutions, and evaluating a subset of solutions in a manufacturing case study. We utilize a CNC machining dataset with over 100,000 data points per time-series. We examine algorithms representing two main approaches: feature engineering for data transformation and dimensionality reduction, and advanced machine learning and deep learning models to capture long-term dependencies. Specifically, we implement Discrete Fourier Transform, Symbolic Aggregate Approximation, Symbolic Fourier Approximation, ARSENAL, DrCIF, and ResNet algorithms for TSC. Experimental results highlight the trade-off between accuracy and computational expense. While ARSENAL and DrCIF achieve high accuracy, their runtimes are notably high. Feature engineering methods offer competitive accuracy with very low runtimes, and ResNet strikes a compelling balance by surpassing benchmarks in accuracy while maintaining reasonable computational costs. These findings can aid in adopting appropriate tools to successfully harness manufacturing big data to derive actionable insights for process improvements and optimization. The novelty of this research lies in addressing the specific challenges of classification tasks on high-resolution time-series data within the manufacturing domain, exploring diverse techniques, applying them to a case study, and providing a nuanced analysis of algorithmic performance with practical implications.

On the Value of Informal Communication in Archaeological Data Work

Abstract Archaeological data simultaneously serve as formal documentary evidence that supports and legitimizes chains of analytical inference and as communicative media that bind together scholarly activities distributed across time, place, and social circumstance. This contributes to a sense of “epistemic anxiety,” whereby archaeologists require that data be objective and decisive to support computational analysis but also intuitively understand data to be subjective and situated based on their own experiences as participants in an archaeological community of practice. In this article, I present observations of and elicitations about archaeological practices relating to the constitution and transformation of data in three cases in order to articulate this tension and document how archaeologists cope with it. I found that archaeologists rely on a wide variety of situated representations of archaeological experiences – which are either not recorded at all or occupy entirely separate and unpublished data streams – to make sense of more formal records. This undervalued information is crucial for ensuring that relatively local, bounded, and private collaborative ties may be extended beyond the scope of a project and, therefore, should be given more attention as we continue to develop open data infrastructures.

STATUS AND DEVELOPMENT TRAJECTORIES CONSUMER AND MORTGAGE LENDING IN THE REPUBLIC OF KAZAKHSTAN

The purpose of the article is to determine the optimal development trajectories based on the research of the mortgage and consumer lending segment in the Republic of Kazakhstan. The mechanism of management of the consumer and mortgage lending system in the system of financial relations between the population and the institution of banking structures; the level of optimal relationships between participants in this process and their regulation through the bank credit rate, investment and marketing activities of the country's banking institutions. The article uses methods of transformation of analytical data display, analysis and synthesis based on qualitative and quantitative approaches. The official statistics of the NBRK and the second-tier banks of the Republic of Kazakhstan were used as the initial information of the analysis. Based on the analysis of the interdependence of mortgage and consumer lending, the level of the base rate of the National Bank of the Republic of Kazakhstan and other factors, recommendations for the development of an optimal trajectory of long-term development are proposed. The problems and prospects of achieving the optimal level of mortgage and consumer lending in the Republic of Kazakhstan, taking into account the interests of both banking institutions and consumers of the country, are considered.

Named entity recognition and its role in unstructured data analysis

In today's digital world, where vast amounts of unstructured data are generated every day, the ability to efficiently process this information is key for many industries. Unstructured data, which includes text files, emails, video, audio, images, and other forms of media, is the bulk of digital data and requires specialized tools to analyze it. Natural Language Processing (NLP) and Named Entity Recognition (NER) are two key technologies that enable the transformation of unstructured data into structured information that can be used for a variety of applications. Natural Language Processing enables machines to understand, interpret, manipulate and generate human language, opening up possibilities for deep analysis of textual data. This includes identifying key words, phrases, themes, and emotional nuances in texts. NER, as an important component of Natural Language Processing, specializes in identifying and classifying named entities in the text into certain categories, such as names of persons, organizations, locations, dates, times, and others. This allows you to automate the processes of sorting, categorizing and analyzing information. However, working with Natural Language Processing and Named Entity Recognition faces a number of challenges. The large volume and variety of data make it difficult to collect, store and analyze it. Lack of standardization can lead to problems with interoperability and integration of different data sources. In addition, there are challenges related to the recognition of named entities, in particular, distinguishing between the same names belonging to different persons and understanding the context in which the names are used. Despite these challenges, the outlook for Natural Language Processing and Named Entity Recognition looks bright, with continued innovations in artificial intelligence and machine learning promising to improve the accuracy and efficiency of these technologies in the future.

Systematic review of Janus kinases inhibitors for rheumatoid arthritis: methodology, reporting, and quality of evidence evaluation.

To evaluate the methodological, reporting and evidence quality of systematic reviews or meta-analyses of Janus kinases (JAK) inhibitors for the treatment of rheumatoid arthritis (RA). Our study systematically retrieved reviews from various databases, spanning from inception to June 2024. Two evaluators independently assessed the methodological, reporting, and evidence quality of each review using the AMSTAR-2 and PRIAMA2020 tools. The evidence quality was evaluated according to GRADE criteria. Six aspects were evaluated: publication year, study type, homogeneity, risk of publication bias, AMSTAR-2 methodology, and PRIAMA2020 reporting quality. Excel 2016 facilitated conversion of scores into radar plots. Following stringent selection criteria, a total of 18 relevant studies were identified. The AMSTAR-2 scores ranged from 4 to 13 points, with five studies rated as low quality and the remaining 13 as critically low quality. All studies encompassed populations, interventions, controls, and outcome measures, demonstrating commendable integrity. However, there is room for improvement in study protocol development and registration, comprehensive search strategies, inclusion and exclusion criteria, conflict of interest disclosure, and discussion of heterogeneity. PRIAMA2020 assessments ranged from 14.5 to 21 points, with two studies scoring below 15 points due to increased bias risk from data transformation and sensitivity analysis. Notably, all reviews (100%) adhered to PRIAMA2020 guidelines for certain items but none met all criteria. GRADE evaluation included 446 outcome measures, with 158 of moderate, 156 of low, and 132 of very low quality, indicating JAK inhibitors is effective in improving RA. According to radar chart, the average rank score was 13.13. One study achieved a balanced score across all dimensions, while 11 exceeded the average, five showed significant differences in PRIAMA2020 scores, and four in AMSTAR two scores. Despite summarizing the efficacy and safety of JAK inhibitors in treating RA, the included studies exhibited poor methodological and reporting quality, along with low-quality evidence overall. Therefore, caution is warranted among decision-makers regarding the use of JAK inhibitors in RA treatment. Urgent requirements include high-quality, multicenter studies investigating JAK inhibitors for RA. https://www.crd.york.ac.uk/PROSPERO, identifier 413415.

Implementasi Algoritma K-Means Guna Pengelompokkan Data Penjualan Berdasarkan Pembelian

Information technology can help solve problems faced by humans by facilitating performance. Information technology and information systems are difficult to separate in the business world. Data mining is the core of the KDD process, which involves inferring algorithms that explore data, developing models and finding previously unknown patterns. KDD is often referred to as knowledge discovery in databases. The KDD process generally consists of 5 stages, namely data selection, pre-processing/cleaning, transformation, data mining and interpretation/evaluation. K-Means is a clustering algorithm in data mining to be able to produce groups of large amounts of data with a point-based partition method with fast and efficient computing time. Clustering is the process of dividing objects from a data set into several homogeneous clusters. The main purpose of the cluster method is to group a number of data/objects into clusters (groups) so that each cluster will contain data that is as similar as possible. This study aims to provide real solutions to UD. Martua in order to know which items are selling well and which items are not selling well so that the object can know which items need to be added to the stock and which items need to be reduced. The method used in this study is the K-Means method with stages, namely data selection, pre-processing, data transformation, information extraction and evaluation of results. The data consists of 30 item data, there are 8 as members of C1 and are best-selling items and 22 as members of C2 and are not selling items. The conclusion that can be obtained from this study is that the K-Means method can group items at UD. Martua. This study shows that the implementation of the K-Means method with the support of the RapidMiner application is effective in grouping item data at UD. Martua.

Creation of an Automated and Comprehensive Resident Progress System for Residents and to Save Hours of Faculty Time: Mixed Methods Study.

It is vital for residents to have a longitudinal view of their educational progression, and it is crucial for the medical education team to have a clear way to track resident progress over time. Current tools for aggregating resident data are difficult to use and do not provide a comprehensive way to evaluate and display resident educational advancement. This study aims to describe the creation and assessment of a system designed to improve the longitudinal presentation, quality, and synthesis of educational progress for trainees. We created a new system for residency progress management with 3 goals in mind, that are (1) a long-term and centralized location for residency education data, (2) a clear and intuitive interface that is easy to access for both the residents and faculty involved in medical education, and (3) automated data input, transformation, and analysis. We present evaluations regarding whether residents find the system useful, and whether faculty like the system and perceive that it helps them save time with administrative duties. The system was created using a suite of Google Workspace tools including Forms, Sheets, Gmail, and a collection of Apps Scripts triggered at various times and events. To assess whether the system had an effect on the residents, we surveyed and asked them to self-report on how often they accessed the system and interviewed them as to whether they found it useful. To understand what the faculty thought of the system, we conducted a 14-person focus group and asked the faculty to self-report their time spent preparing for residency progress meetings before and after the system debut. The system went live in February 2022 as a quality improvement project, evolving through multiple iterations of feedback. The authors found that the system was accessed differently by different postgraduate years (PGY), with the most usage reported in the PGY1 class (weekly), and the least amount of usage in the PGY3 class (once or twice). However, all of the residents reported finding the system useful, specifically for aggregating all of their evaluations in the same place. Faculty members felt that the system enabled a more high-quality biannual clinical competency committee meeting and they reported a combined time savings of 8 hours in preparation for each clinical competency committee as a result of reviewing resident data through the system. Our study reports on the creation of an automated, instantaneous, and comprehensive resident progress management system. The system has been shown to be well-liked by both residents and faculty. Younger PGY classes reported more frequent system usage than older PGY classes. Faculty reported that it helped facilitate more meaningful discussion of training progression and reduced the administrative burden by 8 hours per biannual session.

A comparison of data visualization tools: A case study in health-related research

Data visualization tools are important tools for making health decisions based on large amounts of data. However, very few comparisons of data visualization tools have been done in the context of health-related research. This paper compares commonly available visualization tools from the perspective of a dashboard developer in the context of scientific health research using real-world data and to provide general suggestions for dashboard developers’ future visualization efforts in health-related scientific discoveries. We evaluated four commonly available visualization tools (Tableau, Spotfire, PowerBI, and R Shiny). Each tool was evaluated on visualization outcomes (comparing the generated outputs of tables, scatterplots, correlation plots, treemaps, heatmaps, and geographic maps), usability, Cloud compatibility, analytic integration, and a few other factors. All tools generated comparable visualization outcomes in our tests and were equally usable with the exception of R Shiny which is not recommended for users with limited coding experience. Every tool except R Shiny supports connections to PostgreSQL, gSheets, and Athena without additional configurations. All tools support integration with R and Python; however, PowerBI requires some additional configuration. Every tool is capable of data pre-processing and supports data transformation, null detection and outlier detection. Each tool has its own unique advantages and limitations.

MolPipeline: A Python Package for Processing Molecules with RDKit in Scikit-learn.

The open-source package scikit-learn provides various machine learning algorithms and data processing tools, including the Pipeline class, which allows users to prepend custom data transformation steps to the machine learning model. We introduce the MolPipeline package, which extends this concept to cheminformatics by wrapping standard RDKit functionality, such as reading and writing SMILES strings or calculating molecular descriptors from a molecule object. We aimed to build an easy-to-use Python package to create completely automated end-to-end pipelines that scale to large data sets. Particular emphasis was put on handling erroneous instances, where resolution would require manual intervention in default pipelines. MolPipeline provides the building blocks to enable seamless integration of common cheminformatics tasks within scikit-learn's pipeline framework, such as scaffold splits and molecular standardization, making pipeline building easily adaptable to diverse project requirements.

PDSMV3-DCRNN: A novel ensemble deep learning framework for enhancing phishing detection and URL extraction

Phishing is a cyber-attack that exploits victims' technical ignorance or naivety and commonly involves a Uniform Resources Locator (URL). As a result, it is beneficial to examine URLs before accessing them to spot a phishing assault. Several algorithms based on machine learning have been presented to detect phishing attempts. However, these approaches often suffer from lower performance outcomes, such as lower accuracy, longer response times, and higher false positive rates. Furthermore, many existing methods rely heavily on predefined feature sets, which may limit their adaptability and robustness. In contrast, our proposed method leverages a more dynamic feature selection process, which includes the Conditional Wasserstein Generative Adversarial Network (CWGAN) for addressing data imbalance and the Binary Grey Goose Optimization Algorithm (BGGOA) for optimal feature selection. This dynamic approach enhances the model's ability to adapt to varying data characteristics, improving detection performance. The proposed solution is divided into two stages: pre-deployment and deployment. During the pre-deployment stage, the dataset is preprocessed, including data transformation, handling irrelevant and redundant data, and ensuring data balancing. Minority samples are increased using CWGAN to avoid class imbalance. Features are then selected using BGGOA, resulting in a feature-reduced dataset used for training and testing ensemble deep learning classifiers, specifically the Novel Pyramid Depth-wise Separable-MobileNetV3 (PyDS-MV3) and Deformable Convolutional Residual Neural Network (DCRNN), termed PDSMV3-DCRNN. During the deployment phase, the Boosted ConvNeXt approach extracts URL features fed into the trained classifier to predict "phishing" or "benign". According to experimental findings, the proposed solution outperforms all other tested approaches, displaying a faster training time of 0.11 s and achieving an optimal accuracy of 99.21%.