Network Data Research Articles

DGTNet：dynamic graph attention transformer network for traffic flow forecasting

Abstract Graph-based traffic flow prediction plays a crucial role in urban traffic management and planning. In this paper, we propose a novel Dynamic Graph Attention Transformer Network (DGTnet), which is designed to address the issue of inadequately integrating of temporal and spatial dimensions in traditional models. DGTnet maintains temporal continuity while revealing the complex dynamic relationships between key nodes in the urban traffic system, capturing the periodic changes in the rhythm of city life. Specifically, this study adopts adaptive signal decomposition technology to decompose traffic data into multiple intrinsic mode functions (IMFs), effectively capturing the dynamic changes in traffic flow. This decomposition method is key to the implementation of DGTnet dynamic graph construction, enabling the analysis of traffic flow at different time scales, thereby providing a new perspective for traffic flow prediction research. In the traffic prediction module, we comprehensively consider node, edge, and graph structural information, adopting a multi-head self-attention mechanism to achieve direct cross-modeling in both temporal and spatial dimensions. Finally, we introduce a position-wise feed forward network layer to integrate different types of data and capture nonlinear relationships. The experimental results, based on the public transportation network data sets METR_LA, PEMS_BAY, PEMS03 and PEMS07, demonstrate that DGTNet exhibits notable enhancements in three evaluation indicators, namely the Mean Absolute Percentage Error (MAPE), the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE). The pertinent code has been made available for public access at https://github.com/chenjing0616/DGTNet.

Local and translocal social capital in flood adaptation: evidence from Indonesian coastal communities

ABSTRACT Environmental change and sea level rise challenge urban coastal communities, especially in the Global South. Existing hazard research emphasises the vital role of social capital in adaptation but often overlooks translocal social capital, which encompasses social support and resources that extend beyond local boundaries and connect people at different places. To reduce this research gap, this study investigates the impact of local and translocal social capital on flood-prone households in Padang and Denpasar, Indonesia, using survey data (N = 620) and social network data (N = 1169). The results show translocal contacts provide more emotional, moral, financial, and material support, while local contacts share flood-related knowledge, skills, advice, and practical support. Further, our findings show that the effectiveness of social support varies: local support is strengthened by community ties and diverse forms of support, whereas translocal support's significant role in mental health relies on intimacy and reliability in strong personal relationships. Well-connected support contacts help households to access broader social networks. These findings highlight the different roles of local and translocal social capital in long-term adaptation, which partly complement each other. Households benefit from maintaining diverse support networks, with each type of social capital playing distinct yet interdependent roles in enhancing resilience against environmental changes.

Mining teacher informal online learning networks: Community commitment in unstructured learning environments

AbstractBackgroundSocial media provides new opportunities for teachers to learn, communicate and develop professional relationships. It has been proved to be a valid and helpful resource for teachers' professional learning purposes.ObjectivesWhile previous studies pursued questions like how participants feel, how to support interaction and why participants remain committed, we asked a more fundamental question: what is the structure of a massive informal online professional learning network and what dynamics can we expect regarding participants' commitment?MethodsThis work presents an empirical study of massive informal online professional networks to investigate the dynamics of learning communities and participants' commitment over time. We employed social network analysis and data mining techniques on a longitudinal data set of more than 400,000 tweets published with the hashtag “#edchat.”Results and ConclusionsWe found that around 30% of participants remained committed to the informal learning community over time. Meanwhile, as more and more people committed to the online learning community, participants tended to form smaller communities where the internal connection was stronger.TakeawaysIn informal online learning environments, participants can form stable connections. The 30% threshold can be used to measure massive informal online learning networks in terms of commitment or persistence.

Timely Detection of DDoS Attacks with Dimenstionality Reduction

Due to the interconnectedness and exponential proliferation of IoT devices, the technology is more susceptible to network attacks like Distributed Denial of Service (DDoS), which disrupt network resources. A growing threat to cloud computing systems is the Distributed Denial of Service (DDoS) attack, in which the attacker starts the attack by taking advantage of computers both inside and outside the cloud system. Real-time analysis of cloud network data is essential for preventing DDoS attacks. DDoS attacks interfere with the operation of Io- connected apps and services by taking advantage of the constrained resources on IoT devices. The impacts of DDoS attacks, which seriously damage current systems, are thoroughly examined in this article in the context of the Internet of Things. One of the most common network attacks is the distributed denial-of-service attack (DDoS). DDoS assaults intensified due to the quick development of computer and communication technologies. Therefore, investigating the detection of a DDoS attack is crucial. A single technique cannot offer adequate security due to the variety of DDoS attack techniques.

Efficient Workflow Scheduling for Minimizing Data Transfers and Enhancing Resource Utilization in Cloud IaaS Platforms

Cloud IaaS platforms readily provide access to homogeneous multi-core machines, whether they are physical ("bare metal") or virtual machines. Each of these machines can be equipped with high-performance SSD disks, enabling the distribution of workflow-generated files across multiple machines, which helps minimize the overhead associated with data transfers. In this paper, we propose a scheduling algorithm called SMDT-ERU (Scheduling for Minimizing Data Transfer - Enhancing Resource Utilization), designed to reduce the makespan of data-intensive workflows by minimizing data transfers between dependent tasks over the network. Intermediate files generated by tasks are stored locally on the disk of the machine where the tasks are executed. Through experimentation, we confirm that increasing the number of cores per machine reduces the additional costs caused by network data transfers. Real-world workflow experiments demonstrate the advantages of the proposed algorithm. Our data-driven scheduling approach significantly reduces execution time and the volume of data transferred over the network, outperforming one of the leading state-of-the-art algorithms, which we have adapted to fit our assumptions.

Goals of Surgical Interventions in Youths Receiving Palliative Care

Most youths receiving palliative care undergo many surgical interventions over their lifetimes. The intended purposes of interventions in the context of goals of care are not commonly articulated. To describe the goals and purposes of surgical intervention in youths receiving palliative care and propose a framework discussing intervention using goal-oriented language. This retrospective cohort analysis was conducted among a subset of patients enrolled between April 2017 and March 2021 in a prospective multicenter cohort study of youths receiving palliative care (the Pediatric Palliative Care Research Network's Shared Data and Research [SHARE] Study). Patients younger than 30 years receiving palliative care services were eligible for inclusion in SHARE, and all enrolled at Boston Children's Hospital/Dana Farber Cancer Institute, a SHARE site, were included in this study. Goals and purposes of all surgical interventions from the time of diagnosis through the present were abstracted from patient records. A goal and purpose framework was generated using a hybrid deductive-inductive approach based on established goals-of-care frameworks and the clinical context of surgical interventions. Data were analyzed in September 2023. Primary outcomes included goals and purposes of surgical interventions performed in the study population. Among 197 youths receiving palliative care (mean [SD] age at palliative care start, 8.01 [7.53] years; 108 male [54.8%]; 6 Asian [3.0%], 12 Black [6.1%], 129 White [65.5%], and 16 with >1 race [8.1%]; 27 Hispanic [13.7%] and 142 not Hispanic [72.1%]), almost all individuals (189 youths [95.9%]) underwent at least 1 surgical intervention (mean [SD] 17.5 [16.3] interventions; median [IQR] 13 [5-22] interventions). Of 3331 surgical interventions, there were 878 interventions (26.5%) conducted with the goal of life extension, 1229 interventions (37.1%) conducted for life enhancement, and 79 interventions (2.4%) conducted for both goals; the remaining 1130 interventions (34.1%) held neither goal. Most interventions were performed with the purpose of diagnosis (1092 interventions [32.9%]) or cure and repair (1055 interventions [31.8%]), with fewer performed for the purpose of placing or maintaining assistive technology (696 interventions [21.0%]) or for supportive (434 interventions [13.1%]) or temporizing (39 interventions [1.2%]) purposes. Patients with cardiovascular disease and cancers constituted approximately half (592 patients [56.1%]) of those undergoing curative or repair interventions, whereas youths with neurologic or genetic conditions constituted approximately half (244 patients [56.2%]) of those undergoing supportive interventions. In this cohort study, nearly all youths underwent surgical intervention, and the purposes of intervention differed by serious illness type. These findings suggest that conversations centered on a proposed framework concerning goals and purposes of surgical intervention may facilitate goal-concordant, high-quality care for youths with serious illness.

ML-based Anomaly Detection for CAN Bus Network in Agriculture Machinery

The adoption of advanced automation and next-generation technologies like the Internet of Things (IoT) and modern communication networks has revolutionized the food and agriculture sector, boosting the efficiency and precision of farm machinery. However, this increased inter-connectivity has also exposed significant vulnerabilities, particularly in Controller Area Network (CAN) protocols, widely used in advanced agricultural machinery and equipment. Due to its lack of inherent security features, CAN is susceptible to various cyber-attacks, potentially leading to severe consequences if these attacks remain undetected and unmitigated. This paper introduces a supervised machine learning (ML)-based anomaly detection system (CAN-ADS) designed to detect various cyber-attacks on CAN-based agricultural machinery. The system leverages network traffic augmentation and data balancing techniques to train ML algorithms on CAN-specific datasets. Experimental results show that CAN-ADS achieves high accuracy (approx 98%) and true-positive rates with low false-negative rates (approx 1%).

Application of convolutional neural networks in image classification and applications of improved convolutional neural networks

Abstract. This paper reviews the application and improvement of convolutional neural networks (CNNs) in image classification. Firstly, a shallow CNN for interstitial lung disease image classification is presented. This model suppresses overfitting through a unique network architecture and optimisation algorithm. Next, the improved VGG16 architecture and MIDNet18 model are discussed and their superior performance in brain tumour image classification is demonstrated. Subsequently, a CNN-CapsNet model for cervical cancer image classification and its improvement are presented and the customised model is compared with the conventional VGG-16 CNN architecture in the paper. Next, the application of sparse convolutional kernels and hybrid sparse convolutional kernels (HDCs) in solving the problem of computational resource consumption is presented. Subsequently, methods for solving the problem of limited training data through transfer learning and network data augmentation techniques are discussed, as well as GAN-generated datasets for solving the overfitting problem. Finally, the effect of degraded images on the classification effectiveness of CNNs is explored. The results show that the improved CNN architecture and algorithms have significant effects in solving the problems of overfitting and computational resource consumption, and can significantly improve the accuracy and efficiency of image classification. And degraded images do adversely affect the accuracy of CNN for image classification.

Enhanced Short-Term Temperature Prediction of Seasonally Frozen Soil Subgrades Using the NARX Neural Network

Accurate prediction of subgrade temperatures in seasonally frozen regions is crucial for understanding thermal states, frost heave phenomena, stability, and other critical characteristics. This study employs a nonlinear autoregressive with exogenous input (NARX) network to predict short-term subgrade temperatures in the Golmud-Nagqu section of China’s National Highway 109. The methodology involves preprocessing subgrade monitoring data, including temperature, water content, and frost heave, followed by developing a temperature prediction model. This tailored NARX neural network, compared to the traditional BP neural network, integrates feedback and delay mechanisms for monitoring data, offering superior memory and dynamic response capabilities. The precision of the NARX model is assessed with the backpropagation (BP) network, indicating that the NARX neural network significantly outperforms the BP model in both precision and stability for temperature prediction in seasonally frozen subgrades. These findings suggest that the NARX model is a valuable tool for accurately predicting subgrade temperatures in seasonally frozen regions, offering significant insights for practical engineering applications.

FusionGCN: Multi-focus image fusion using superpixel features generation GCN and pixel-level feature reconstruction CNN

In recent years, convolutional neural networks have demonstrated significant advancements in the domain of computer vision, effectively addressing numerous previously challenging issues. An increasing number of researchers are focusing their investigations on this field, proposing innovative network architectures. However, many existing networks necessitate intricate module designs and a substantial number of parameters to achieve satisfactory fusion outcomes, which poses challenges for lightweight devices with constrained computational resources. To mitigate this concern, the present study introduces a novel methodology that integrates block segmentation with pixel optimization. Specifically, we initially employ graph convolutional networks to execute flexible convolutions on large-scale, irregular regions generated through superpixel clustering, thereby achieving coarse segmentation at the block level. Subsequently, we utilize parallel lightweight convolutional networks to provide pixel-level guidance, ultimately resulting in a more accurate decision map. Furthermore, to leverage the strengths of both networks and facilitate the optimization of feature generation from the graph convolutional network for non-Euclidean data, we meticulously design a superpixel-based graph decoder alongside a pixel-based convolutional neural network extraction block to enhance feature acquisition and propagation. In comparison to numerous state-of-the-art methodologies, our approach demonstrates commendable competitiveness in both qualitative and quantitative analyses, as well as in efficiency evaluations. The code can be downloaded at https://github.com/ouyangbaicai/FusionGCN.

Research on University Network Data Anomaly Detection and Security Protection Algorithm Based on edge computing

Research on University Network Data Anomaly Detection and Security Protection Algorithm Based on edge computing

Photonic quantum generative adversarial networks for classical data

Photonic quantum generative adversarial networks for classical data

Augmented Reality Adoption: Exploring Security Concerns and Consumer Resourcefulness in the Era of Society 5.0

ABSTRACTSociety 5.0, building its foundation on industry 4.0, defines the concept of super smart cities as places where humans can adapt novel technologies such as augmented reality (AR) in their daily lives. AR brought its consumers an immersive and hybrid experience, blurring the boundaries between the real and virtual worlds. This study explores the user adoption of AR, an interactive technology, in the e‐commerce domain in the present era of Society 5.0. This study seeks to address the question of what factors drive the consumer's adoption of AR while shopping online and how security concern and technostress (TS) impact their intention towards its use and its actual usage behaviour. This research offers a novel perspective and fills a gap in the existing literature by integrating consumers' security concerns and TS, previously unexplored factors, in the Unified Theory of Acceptance and Use of Technology (UTAUT) model. The ability of the UTAUT model, over the earlier technology adoption models, to incorporate context‐specific factors makes it significantly relevant for studying AR adoption, a technology blending the virtual and real world. A modified UTAUT‐based questionnaire assessed relevant factors influencing this behaviour. This study employed a quantitative, cross‐sectional survey design to apprehend a broad range of users and to investigate consumer perceptions and adoption of AR in online shopping. Purposive sampling ensured that respondents were aligned with the research context and were knowledgeable about the use of AR in the e‐commerce domain and its relevant applications, thereby strengthening the validity of the study findings. A two‐staged partial least square structural equation modelling–artificial neural network (PLS‐SEM–ANN) data analysis approach was used to determine the study's findings. The meaningful responses collected from 457 respondents were analysed using this targeted approach for in‐depth insights. The synthesis of the study's analysis suggests that personal innovativeness towards information technology and security concerns are critical drivers of intention and actual usage. Interestingly, TS and social influence emerged as non‐significant, indicating a nuanced view of the UTAUT model in this context. A possible explanation for their insignificance lies in the present generation's ability to rapidly adapt to new technologies and make autonomous decisions, reducing reliance on peer influence. By understanding these factors, the research seeks to pave the way for the efficacious integration of AR within the e‐commerce domain. It further emphasizes the significance of consumer resourcefulness, a gap in current research, for bridging the intention‐usage gap. This focus on consumers’ personal innovativeness towards IT, security concerns and consumer resourcefulness provides valuable insights for e‐commerce platforms seeking to drive successful AR adoption. The study explores how well AR will be accepted and used, informing scholarly researchers and businesses on making AR‐integrated online shopping successful.

How Clinician-Scientists Access and Mobilise Social Capital and Thus Contribute to the Professional Development of Their Colleagues in Their Networks

ABSTRACT Clinician-scientists, physicians who conduct research, may fulfil a bridging role in networks of health care researchers and practitioners. Within clinician-scientists’ networks, knowledge sharing is thought to play a vital role in the continuing professional development of themselves and their colleagues. However, little is known about networks of clinician-scientists and how this impacts continuing professional development. Rooted in social capital theory, this study provides a mixed methods exploration of clinician-scientists’ networks. Ego-level social network data were collected via semi-structured interviews on professional interactions about evidence-based practice with 15 clinician-scientists in the area of general practice and elderly care. Quantitative analysis revealed that professional networks of clinician-scientists varied in size, composition, and frequency of interactions depending on appointed research time and experience. Less experienced clinician-scientists interacted most frequently with other clinician-scientists while experienced clinician-scientist experienced more sporadically with clinicians. Clinician-scientists with more research time interacted more frequently with scientists and had a slightly larger professional network than those with less research time. The thematic qualitative analysis revealed different decision-making processes of clinician-scientists on mobilising their social capital and connecting to others in the network: (1) deliberate decision about initiating connections; (2) reactive behaviour without a decision; (3) ad-hoc decision. Clinician-scientists exchange knowledge to enhance their own continuing professional development mainly but also contribute to the professional development of clinicians, scientists, and other clinician-scientists.

LEAF: A Lifestyle Approximation Framework Based on Analysis of Mobile Network Data in Smart Cities

The increasing use of mobile networks is an opportunity to collect and model users’ movement data for extracting knowledge about life and health while considering privacy leakage risk. This study aims to approximate the lifestyles of urban residents, employing statistical information derived from their movements among various Points of Interest (PoI). Our investigations comprehend a multidimensional analysis of key urban factors to provide insights into the population’s daily routines, preferences, and characteristics. To this end, we developed a framework called LEAF that models lifestyles by interpreting anonymized cell phone mobility data and integrating it with information from other sources, such as geographical layers of land use and sets of PoI. LEAF presents the information in a vector space model capable of responding to spatial queries about lifestyle. We also developed a consolidated lifestyle pattern framework to systematically identify and analyze the dominant activity patterns in different urban areas. To evaluate the effectiveness of the proposed framework, we tested it on movement data from individuals in a medium-sized city and compared the results with information collected through surveys. The RMSE of 5.167 between the proposed framework’s results and survey-based data indicates that the framework provides a reliable estimation of lifestyle patterns across diverse urban areas. Additionally, summarized patterns of criteria ordering were created, offering a concise and intuitive representation of lifestyles. The analysis revealed high consistency between the two methods in the derived patterns, underscoring the framework’s robustness and accuracy in modeling urban lifestyle dynamics.

Power flow distribution identification via machine learning algorithm and distributed resource clearing method

Abstract Under the background of dual carbon, a large number of distributed new energy sources are connected to the demand side, which leads to great risks in the operation of the power system. As an effective solution, distributed power transaction still has problems, such as high clearing time, unknown low-voltage distribution network structure, and resource scheduling ignoring user needs. Therefore, this paper proposes an optimal regulation method of demand-side resources based on distributed transactions. Based on the historical operation data of the distribution network, the eXtreme Gradient Boosting (XGBoost) algorithm is used to study the relationship between node injection power, node voltage, and node-side line power flow. It is also used to check the results of distributed transactions and design a market mechanism for rapid clearing and settlement of distributed transactions. The results of an example show that the identification of power flow distribution in a distribution network based on reinforcement learning can meet the security check requirements of distributed transaction results.

Factors associated with waitlist clinical deterioration among United States lung transplant recipients under the continuous distribution system

BackgroundLung continuous distribution (CD), implemented on March 9, 2023, changed the calculation and relative importance of medical urgency and posttransplant survival in prioritizing candidates for transplant. We aimed to identify factors associated with waitlist clinical deterioration and change in expected posttransplant survival from listing to transplant in the current system. MethodsWe used Organ Procurement and Transplantation Network (OPTN) data to conduct a retrospective study of 2,395 adult, lung-only transplant recipients added to the waiting list and transplanted between March 09, 2023, and March 08, 2024. We used multivariable linear regression to identify factors associated with change in waitlist area under the curve (WLAUC) and posttransplant area under the curve (PTAUC), representing expected survival (in days) without and with transplant, respectively. ResultsIn multivariable analysis, longer waiting time (β=-1.3 per 7 days; p<0.001), male birth sex (β=-11; p=0.006), diagnosis group D (β=-27; p<0.001), and blood type O (β=-13; p<0.001) were associated with greater clinical deterioration from listing to transplant. Older (β=3.2 per 10-year increase in age; p=0.023) and taller (β=3.0 per 5 cm increase in height; p=0.003) recipients were less likely to clinically deteriorate from listing to transplant. Diagnosis group D (β=-4.7, p=0.032) and blood type O (β=-4.2, p=0.025) recipients had lower expected posttransplant survival at transplant compared to listing. ConclusionOur findings suggest the need to further investigate and address factors associated with waitlist clinical deterioration under CD. Future analysis of the effect of waitlist clinical deterioration on posttransplant outcomes under CD is needed.

Rapid deep learning prediction model using satellite imagery for radiation accident Announcement system in Serbia

Radioactivity environmental monitoring with the help of the Radiation Accident Announcement System (RAAS) established in the Republic of Serbia is of vital importance for rapid response in the event of intervention dose values being reached such as Operational Intervention levels (OILs). There are cases of impossibility of using a ground-based model in order to predict the transport and spread of radiation during a nuclear accident such as the one in the Fukushima Daiichi nuclear power plant 2011. Modern technology has made it possible to use machine learning models with remote sensing data in addition (or an alternative) to atmospheric models with ground data collection methods. Deep learning (DL) model was developed and trained on a satellite-based cloud fraction dataset to forecast diurnal cloud drift. By forecasting the movement of clouds that are potential carriers of radioactive materials, decision-makers can provide an adequate response with an Action Plan in the case of nuclear and radiation accidents and incidents. Designed Application Programming Interface (API) has been developed and integrated between the DL model and the RAAS. In this way, the monitoring, data sharing and exchange processes were automated in order to trigger the DL model when an OILs value of 1 μSv/h is reached. The hyperparameter optimization process of the DL model was done with grid search and Particle Swarm Optimization (PSO) to achieve maximum performance on the data in a reasonable amount of time. The evolution metrics to monitor and measure the performance of the DL model were cosine similarity (CS) and structural similarity (SS) with the best score of 0.78282 and 0.27063 for grid search, respectively, while 0.27065 and 0.78282 for PSO, respectively. It can be concluded that remote sensing imagery with DL model is an alternative approach against a ground-based forecasting system, and is able to predict in near real-time independently of ground network system and data.

Lymph node metastasis prediction from in-situ lung squamous cell carcinoma histopathology images using deep learning.

Lung squamous cell carcinoma (LUSC), a subtype of non-small cell lung cancer, represents a significant portion of lung cancer cases with distinct histologic patterns impacting prognosis and treatment. The current pathological assessment methods face limitations such as inter-observer variability, necessitating more reliable techniques. This study seeks to predict lymph node metastasis in LUSC using deep learning models applied to histopathology images of primary tumors, offering a more accurate and objective method for diagnosis and prognosis.Whole slide images (WSIs) from the Outdo-LUSC and TCGA-LUSC cohorts were used to train and validate deep-learning models. Multi-instance learning was applied, with patch-level predictions aggregated into WSI-level outcomes. The study employed the ResNet-18 network, transfer learning, and rigorous data preprocessing. To represent WSI features, innovative techniques like patch likelihood histogram (PLH) and bag of words (BoW) were used, followed by training of machine learning classifiers, including the ExtraTrees algorithm.The diagnostic model for lymph node metastasis showed strong performance, particularly using the ExtraTrees algorithm, as demonstrated by receiver operating characteristic (ROC) curves and Grad-CAM visualizations. The signature generated by the ExtraTrees algorithm, named LN_ISLUSCH (lymph node status-related in-situ lung squamous cell carcinoma histopathology), achieved an area under the curve (AUC) of 0.941 (95% CI: 0.926-0.955) in the training set and 0.788 (95% CI: 0.748-0.827) in the test set. Kaplan–Meier analyses confirmed that the LN_ISLUSCH model was a significant prognostic factor (p = 0.02).This study underscores the potential of artificial intelligence in enhancing diagnostic precision in pathology. The LN_ISLUSCH model stands out as a promising tool for predicting lymph node metastasis and prognosis in LUSC. Future studies should focus on larger and more diverse cohorts and explore the integration of additional omics data to further refine predictive accuracy and clinical utility.

A protocol for trustworthy EEG decoding with neural networks

Deep learning solutions have rapidly emerged for EEG decoding, achieving state-of-the-art performance on a variety of decoding tasks. Despite their high performance, existing solutions do not fully address the challenge posed by the introduction of many hyperparameters, defining data pre-processing, network architecture, network training, and data augmentation. Automatic hyperparameter search is rarely performed and limited to network-related hyperparameters. Moreover, pipelines are highly sensitive to performance fluctuations due to random initialization, hindering their reliability. Here, we design a comprehensive protocol for EEG decoding that explores the hyperparameters characterizing the entire pipeline and that includes multi-seed initialization for providing robust performance estimates. Our protocol is validated on 9 datasets about motor imagery, P300, SSVEP, including 204 participants and 26 recording sessions, and on different deep learning models. We accompany our protocol with extensive experiments on the main aspects influencing it, such as the number of participants used for hyperparameter search, the split into sequential simpler searches (multi-step search), the use of informed vs. non-informed search algorithms, and the number of random seeds for obtaining stable performance. The best protocol included 2-step hyperparameter search via an informed search algorithm, with the final training and evaluation performed using 10 random initializations. The optimal trade-off between performance and computational time was achieved by using a subset of 3-5 participants for hyperparameter search. Our protocol consistently outperformed baseline state-of-the-art pipelines, widely across datasets and models, and could represent a standard approach for neuroscientists for decoding EEG in a trustworthy and reliable way.