Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
Vision-Based Aerial Navigation Using Satellite Offline Maps
No full textWhile Global Navigation Satellite System (GNSS) technology combined with advanced augmentation techniques provides high accuracy, its reliance on signal integrity leaves safety-critical applications vulnerable to spoofing and jamming. To address this, a lightweight vision-based localization module is proposed as a self-contained GNSS substitute for small Uncrewed Aerial Vehicles (UAVs). It leverages offline satellite maps and an ensemble of Convolutional Neural Network (CNN) models to estimate 2D offsets and associated uncertainty for stable EKF fusion. The system is designed for seamless integration with the PX4 Autopilot and real-time operation on resource-limited platforms. Software In The Loop (SITL) simulations over varied mission scenarios at Virginia Tech's Kentland Farm demonstrate an R95 accuracy of 4.5 m to 5.5 m, with effective drift bounding even in low-texture environments. Remaining challenges include uncertainty calibration, particularly over tree-covered regions, before transitioning to field testing.Master of ScienceModern drones often rely on satellite navigation, but these signals can be jammed or spoofed, creating risks for safety-critical missions. To reduce this dependence, this work proposes a lightweight, vision-based localization module that can act as a self-contained replacement for satellite navigation on small Uncrewed Aerial Vehicles (UAVs). The system uses pre-downloaded satellite maps and Artificial Intelligence (AI) to estimate how far the UAV has moved and how confident it is in that estimate. This allows the module to integrate smoothly with standard flight-control software like the PX4 Autopilot and run in real time on limited onboard hardware. In simulation tests across varied mission scenarios at Virginia Tech's Kentland Farm, the approach typically keeps the UAV's estimated position within about 4.5 to 5.5 meters. It also limits long-term drift, even when the ground has few visual features. The main remaining challenge is improving how the system measures and reports its own uncertainty, especially over tree-covered areas, before moving from simulation to field testing
Toward Sustainable Digital Infrastructure: Thermal and Economic Potential of Data Center Heat Reuse
Full text linkData centers consume large amounts of electricity, with a significant fraction dissipated as low grade heat. Generally, this waste heat is released to the environment, but growing energy demands and decarbonization targets have emphasized the interest in its recovery and reuse. This study investigates the initial technoeconomic viability of heat reuse from a modeled system with varying effeciency and other key influencing parameters. By using the thermal approximations, we quantify the available temperature ranges and heat flows for several reuse pathways, including district heating, absorption cooling, and domestic hot water production. The results will reflect on the key parameters including the thermal efficiency of direct reuse of heat for low temperature applications, while higher temperature uses necessitate auxiliary upgrades that reduce overall system performance. A comparative techno-economic analysis highlights the trade-offs between capital investment, operating cost, and key data center parameters across reuse scenarios. In particular, coupling with district heating networks emerges can be one of the most scalable option, though decentralized applications (e.g., building heating) can offer faster payback under certain operating regimes. The findings underscore that while data center heat reuse is technically feasible, its practical deployment depends strongly on local infrastructure compatibility and economic drivers. This work contributes a structured framework for evaluating reuse pathways, providing both performance metrics and cost considerations to guide decision making for sustainable data center operation.Published versionYes, full paper (Peer reviewed?
Scenario-Based Methodology for Predicting the Safety Benefits of Emerging Advanced Rider Assistance Systems (ARAS)
No full textMotorcycle crashes remain a critical public health challenge in the United States (U.S.). Despite decades of progress in automotive safety, motorcyclists continue to experience disproportionately high rates of serious and fatal injuries. The combination of inherent vulnerability, limited physical protection, and exposure to complex traffic environments led the motorcycle fatality rate per registered vehicle in the U.S. to remain almost constant over the last forty years.
This increasing concern has elevated interest in emerging Advanced Rider Assistance Systems (ARAS), which aim to reduce both motorcycle crash occurrence and riders' injury severity. However, unlike more mature safety technologies available for passenger vehicles, most motorcycle-specific ARAS remain in early development stages, and their real-world effectiveness is still largely unknown. Progress toward understanding their potential benefits is further constrained by motorcycle-specific limitations, such as the limited crash data available and the absence of a standardized framework for evaluating and comparing system benefits.
To address these gaps, this dissertation presents a novel framework serving as the scientific foundation for ARAS benefit assessment, using Motorcycle Autonomous Emergency Braking (MAEB) as a proving case. Central to this work is a novel scenario-based methodology—adapted from automated-driving research—that integrates national crash statistics, in-depth crash reconstructions, naturalistic riding behavior, and computer-based simulation to produce the first U.S.-based ARAS safety-benefit estimation.
Collectively, this work identifies critical opportunities to improve motorcycle safety through refined ARAS design, scenario-based evaluation, and human-aware control strategies. The scenario-based methodology developed here provides a data-driven scalable framework that allows manufacturers, regulators, and researchers to estimate safety benefits, refine activation logic, and inform the development of future motorcycle safety systems. Although centered on MAEB, the approach is generalizable to a wide range of emerging ARAS technologies and represents an important step toward reducing motorcycle injuries and fatalities on U.S. roads.Doctor of PhilosophyMotorcycle crashes remain a critical public health challenge in the United States (U.S.). Despite decades of progress in automotive safety, motorcyclists continue to experience disproportionately high rates of serious and fatal injuries. The combination of inherent vulnerability, limited physical protection, and exposure to complex traffic environments led the motorcycle fatality rate per registered vehicle in the U.S. to remain almost constant over the last forty years.
This increasing concern has elevated interest in emerging Advanced Rider Assistance Systems (ARAS), which aim to reduce both motorcycle crash occurrence and riders' injury severity. However, unlike more mature safety technologies available for passenger vehicles, most motorcycle-specific ARAS remain in early development stages, and their real-world effectiveness is still largely unknown. Progress toward understanding their potential benefits is further constrained by motorcycle-specific limitations, such as the limited crash data available and the absence of a standardized framework for evaluating and comparing system benefits.
To address these gaps, this dissertation presents a novel framework serving as the scientific foundation for ARAS benefit assessment, using Motorcycle Autonomous Emergency Braking (MAEB) as a proving case. Central to this work is a novel scenario-based methodology—adapted from automated-driving research—that integrates national crash statistics, in-depth crash reconstructions, naturalistic riding behavior, and computer-based simulation to produce the first U.S.-based ARAS safety-benefit estimation.
Collectively, this work identifies critical opportunities to improve motorcycle safety through refined ARAS design, scenario-based evaluation, and human-aware control strategies. The scenario-based methodology developed here provides a data-driven scalable framework that allows manufacturers, regulators, and researchers to estimate safety benefits, refine activation logic, and inform the development of future motorcycle safety systems. Although centered on MAEB, the approach is generalizable to a wide range of emerging ARAS technologies and represents an important step toward reducing motorcycle injuries and fatalities on U.S. roads
Assessing Extension Engagement and Collaboration Through the Virginia Cooperative Extension Situation Analysis Process
No full textEngagement and collaboration among Extension personnel and community stakeholders are critical for maintaining the relevance, effectiveness, and sustainability of Cooperative Extension programs. This study examined internal and external perceptions of collaboration during the Virginia Cooperative Extension (VCE) Situation Analysis process, which is a formal, statewide assessment conducted every five years to identify and prioritize local community issues. The purpose of this study is to gain additional insight into the perceptions of collaboration between and among employees and stakeholders who participated in the VCE Situation Analysis process and to better understand areas for professional development for employees and stakeholders regarding this process. The objectives of this study were to: (1) describe the perceived strength of collaboration from internal and external perspectives, (2) compare these perceptions across groups, (3) identify differences or similarities in perceived collaborative strength, and (4) determine areas for employee development to enhance future engagement. A modified Wilder Collaboration Factors Inventory (WCFI), excluding the Environment and Resource domains, was administered via Qualtrics to 420 identified participants, yielding 167 responses (39.7% response rate). Quantitative data were analyzed across four collaboration domains: Membership Characteristics, Process/Structure, Communication, and Purpose. Mean ratings across all respondents who completed the survey (n = 102) indicated moderate to strong collaboration (M = 3.71 on a 5-point Likert scale), with the Purpose domain receiving the highest mean (M = 3.78) and Process domain the lowest (M = 3.64). External stakeholders consistently rated collaboration higher (M = 4.10) than internal participants (M = 3.63). The highest-rated indicator was "I have a lot of respect for other people in this collaboration" (M = 4.20), while the lowest was "All of the organizations that we needed to be members of this collaborative group became members" (M = 3.12). Qualitative data from open-ended survey responses (N = 50) and 16 follow-up interviews identified recurring themes of community connection, issue identification, and the need for enhanced communication and engagement strategies. Participants emphasized that stakeholder involvement improves the accuracy of community needs assessments, strengthens trust, and supports the development of relevant, high-impact programs. Key skills identified for effective collaboration included soft skills (listening, communication, facilitation, leadership) and technical competencies (data analysis, process knowledge, time management). Findings suggest that while VCE demonstrates strong collaborative intent and mutual respect, gaps remain in stakeholder inclusion and communication consistency. Improving stakeholder engagement through targeted employee development, intentional outreach, and clearer process communication could strengthen future Situation Analyses. Overall, the study underscores that effective collaboration not only enhances Extension's local relevance but also reinforces accountability and support at federal, state, and local levels.Doctor of PhilosophyThe Virginia Cooperative Extension (VCE) works in every county and city across the Commonwealth to connect communities with education and resources that improve quality of life. Every five years, VCE conducts a process called the Situation Analysis to identify the most important issues facing local communities and to guide Extension's programming and priorities. This process relies on strong collaboration between VCE employees, community members, local government, and other organizations. However, until now, the level of collaboration and engagement in this process has not been formally studied. This study set out to understand how well VCE and its partners are working together during the Situation Analysis and to identify ways to strengthen that collaboration. The research asked both VCE employees and community partners about their experiences through a detailed survey and follow-up interviews. The survey measured how people felt about communication, teamwork, shared purpose, and participation. A total of 102 people completed the survey, representing both Extension staff and community stakeholders. Overall, the results showed that collaboration is a strong part of the VCE Situation Analysis process. Most participants agreed that they respected the people they worked with and clearly understood the goals of the process. However, some areas need improvement. The lowest rated area was membership, meaning not all the organizations that should have been involved were included in the process. Community partners rated the process more positively than VCE employees, suggesting that Extension staff may see more room for improvement from within the organization and face more challenges when having the responsibility to conduct the Situation Analysis. Interviews and open-ended survey questions revealed that collaboration helps build stronger community connections, increases awareness of local issues, and ensures that Extension programs reflect what people truly need. Respondents emphasized the importance of communication, trust, and inclusion. They also identified key skills that make collaboration more successful, such as active listening, clear communication, leadership, organization, and the ability to analyze and use data. The findings show that collaboration and community engagement are essential for VCE's continued success. By improving communication, expanding participation, and helping employees and partners develop both soft skills (listening, communication, facilitation, leadership) and technical competencies (data analysis, process knowledge, time management), Extension can better identify community priorities and deliver more effective programs. Stronger collaboration not only supports Extension's mission but also helps communities across Virginia work together to solve local problems and build a better future
Proposing a Design Theory for a Human-Learning-Guided Virtual Negotiator for Online Trading Platforms
No full textNegotiation-based transactional mechanisms provide flexibility and economic benefits to both sellers and buyers on online trading platforms. Although automated negotiation is a highly desired feature among online platform providers, the complexity and uncertainty of human behavior in human-to-computer (HtC) negotiation make it a problematic solution. This study proposes a design theory for a human-learning guided virtual negotiator (HLG-VN) framework that emulates human learning using multiple machine learning (ML) techniques that collectively mimic four human learning activities: didactic, feedback, observational, and analogical learning. Following the design science research methodology, we built an instantiation system for the proposed design theory and empirically tested it using experiments based on HtC negotiations. The empirical results show that our system outperformed the benchmark system in terms of both economic and some key social-psychological outcomes. Furthermore, the experiment results confirm the effectiveness and correctness of the HLG-VN framework. The proposed design theory provides a theoretical base for using ML techniques to build a virtual negotiator agent for an automated negotiation system. Thus, various agents could be designed and developed based on the theory for online trading platforms, thus improving negotiation efficiency and reducing transaction costs.Accepted versio
Phase-field simulation of freezing water droplet
No full textWhen a water droplet freezes on a cold plate, a pointy tip forms as the result of volume expansion. In this dissertation, we will introduce a quasi-compressible phase-field model that deals with this non-isothermal three-phase system involving water, ice, and air. The water-ice phase transition and the water-air fluid interface are handled by the Allen-Cahn and the Cahn-Hilliard equations, respectively. The governing equations, including the two phase-field equations, the Navier-Stokes equations, and the energy equation, are designed such that the non-negative entropy production is guaranteed. These equations are then solved by finite-element methods using the open-source deal.ii library. Our model reproduces the Gibbs-Thomson and Clausius-Clapeyron equations, which establish the dependence of the melting temperature on interface curvature and pressure, respectively. Furthermore, the built-in quasi-compressibility accurately accounts for the volume change due to the water-ice density contrast during the phase transition. With proper parameters, our simulation captures the pointy tip of the frozen droplet with good agreement with the experiment.Doctor of PhilosophyHave you ever noticed how a drop of water freezes into a shape with a pointy tip when it lands on a cold surface? It's a fascinating process that involves the transformation of water into ice, and it's more complex than it might seem at first glance. In our study, we explore this phenomenon using a special computer model that simulates the freezing process, including the interactions between water, ice, and air.
Our model uses advanced mathematical equations to understand how water turns into ice and how the tiny boundary between water and air behaves during freezing. These equations help us ensure that our simulation follows the laws of nature, specifically the principles of entropy, which in simple terms, measures the disorder or randomness of a system.
To solve these complex equations, we use a powerful computer program. Our simulations are not just theoretical exercises; they accurately predict the formation of the pointy tip on the frozen droplet, much like what we observe in real-life experiments. This success demonstrates the reliability of our model.
Furthermore, our study delves into the intricate details of how the shape of the frozen droplet is influenced by various factors, including the unique conditions at the point where water, ice, and air meet, and how changes in pressure and temperature affect the freezing process.
Our research provides a deeper understanding of the freezing process, which is not only fascinating from a scientific perspective but also has potential applications in various fields, such as climate studies and the development of technologies based on the properties of ice and water
Weed Technology
No full textCotton production in the Texas High Plains faces significant challenges due to water scarcity resulting from uneven rainfall patterns and declining levels of the Ogallala aquifer. Deficit or reduced irrigation is one of the most common water management strategies to increase water use efficiency and cotton productivity in the region. However, deficit irrigation can affect herbicide efficacy on weeds. This study investigates how varying irrigation levels affect herbicide efficacy on weeds in cotton production systems. A two-year field study was conducted at Texas Tech University Quaker Research Farm in 2023 and 2024. The experiment was randomized three times in a split-plot design with two irrigation levels: I1 [100% crop evapotranspiration (ETc) replacement] and I2 [50% ETc replacement] as the main plot factor and different pre-emergent and post-emergent herbicide combinations as the subplot factor. Results indicated that reducing the irrigation level to I2 did not affect the total weed density or biomass production but resulted in decreased Palmer amaranth height and biomass production compared to I1. Among herbicide treatments, acetochlor, prometryn, or smetolachlor PRE fb glyphosate + acetochlor, prometryn, or s-metolachlor POST provided the most effective weed control, reducing total weed density, Palmer amaranth weed density and biomass compared to the untreated control and to PRE alone. Although I2 resulted in lower plant height in both years than I1, it produced comparable cotton biomass and lint yield. Among the herbicide treatments, PRE fb glyphosate + residual herbicide POST yielded significantly higher lint yield than the untreated control in both years. In conclusion, the study demonstrates that deficit irrigation is an effective water conservation technique that maintains cotton yield and herbicide efficacy. Additionally, using PRE fb POST herbicide combinations, farmers can achieve effective weed control and sustain cotton productivity in semi-arid regions.Accepted versio
Cyber-Physical-Social Systems for Autonomous Defense: Enabling Mission-Centric, Adaptive, and Anytime Intelligence
No full textAutonomous Cyber-Physical-Social Systems (CPSSs) are rapidly transforming mission-critical domains such as transportation, defense, emergency response, and smart infrastructure by enabling real-time sensing, decentralized control, and autonomous decision-making. Yet the increasing complexity, interconnectivity, and adversarial exposure of CPSSs make them highly susceptible to evolving cyber threats. Existing security approaches frequently operate in isolation, overlooking the deep interdependencies among cyber, physical, and social components as well as the asymmetric information and strategic dynamics that shape attacker–defender interactions. These gaps highlight unresolved challenges in mission-level risk awareness, adaptive defense under uncertainty, and resource-bounded reasoning.
This dissertation pursues a unified research vision: to develop a mission-centric, uncertainty-aware, and resource-adaptive autonomy framework for resilient CPSS operations.
This vision is realized through three tightly interrelated research thrusts, including Mission Impact Assessment (MIA), Intrusion Response Systems (IRS), and Anytime Inference (AIF), that collectively advance trustworthy decision-making under adversarial, uncertain, and resource-constrained conditions. Rather than functioning as isolated contributions, these tasks form a seamless progression: MIA quantifies mission risk and identifies critical assets, IRS defends those assets adaptively under adversarial uncertainty, and AIF enables both MIA and IRS to operate reliably when computation, information, or time is limited.
Task MIA develops an interdependent Mission Impact Assessment framework, termed iMIA.
By integrating Subjective Bayesian Networks (SBNs) with Subjective Logic-based Hypergame Theory (SL-HGT), iMIA models both epistemic uncertainty and divergent attacker–defender perceptions while capturing interdependencies among assets, services, and tasks. This enables robust mission-outcome inference under missing, noisy, or conflicting information.
A key contribution is the identification of highly critical nodes whose degradation disproportionately impacts mission success, allowing targeted reinforcement strategies to enhance resilience and effectiveness in dynamic threat environments.
Task IRS designs an uncertainty-aware, Deep Reinforcement Learning-based Intrusion Response System for resilient operation in in-vehicle networks. The proposed IRS leverages structured sub-action spaces tailored to attack types and employs entropy regularization to promote robust decision policies under uncertainty. Extensive experiments demonstrate significant reductions in Attack Success Ratio (ASR) and improvements in mission-performance metrics such as route completion and safety compliance. A human-in-the-loop extension further incorporates expert feedback into reward shaping, enabling interpretable, adaptive, and trustworthy defense strategies in safety-critical vehicular environments.
Task AIF introduces an Anytime Inference (AIF) algorithm for SBNs that supports incremental, resource-aware reasoning. The framework employs simulation-based inference, bijective mappings between subjective opinions and probability distributions, and dynamic resource allocation strategies based on entropy and Bayes factor heuristics. The resulting algorithm delivers fast, interruptible, and high-fidelity inference even under tight computational budgets, thus directly enabling responsive mission assessment and adaptive defense in the MIA and IRS tasks. Empirical results show accelerated convergence with increasing sample budgets and the superior scalability of Gibbs sampling in larger networks.
Overall, this dissertation yields three key findings: (i) explicitly modeling epistemic and perceptual uncertainty is essential for achieving high-fidelity mission reasoning and adaptive autonomy; (ii) hypergame-theoretic and uncertainty-aware learning approaches dramatically improve defensive effectiveness in adversarial CPSSs; and (iii) resource-aware anytime inference is critical for timely, trustworthy decision-making in dynamic and constrained environments. Future research may extend this unified framework toward multi-agent CPSS settings, integrate richer forms of human collaboration, and advance scalable, uncertainty-aware learning algorithms for increasingly complex autonomous systems.Doctor of PhilosophyAutonomous systems are increasingly used in critical areas such as transportation, national defense, emergency response, and smart cities. These systems combine software, physical machines, and human interactions to sense their environment, make decisions, and act with little or no human intervention. While this capability brings major benefits, it also introduces serious risks. As these systems become more connected and complex, they are more vulnerable to cyber attacks that can disrupt operations, compromise safety, or cause mission failure.
Most existing security solutions focus on individual components in isolation. For example, protecting software without fully considering how physical devices and human behavior are affected. They also often assume perfect information, even though real-world attackers and defenders operate with uncertainty, incomplete data, and limited time or computing power. As a result, today's approaches struggle to assess mission-level risk, adapt defenses in real time, and make reliable decisions under constraints.
This dissertation addresses these challenges by developing a unified framework for building resilient autonomous systems. The core goal is to enable systems to understand mission risk, defend themselves intelligently, and continue operating even when information, time, or computing resources are limited. This vision is realized through three closely connected research areas: Mission Impact Assessment (MIA), Intrusion Response Systems (IRS), and Anytime Inference (AIF). Together, they form a complete pipeline: first understanding what matters most to the mission, then protecting it against attacks, and finally ensuring decisions remain reliable under tight constraints.
The first part, Mission Impact Assessment, introduces a framework called iMIA that evaluates how cyber attacks affect overall mission success. Rather than focusing on individual failures, iMIA identifies which components are most critical and how their degradation could ripple through the system. By explicitly modeling uncertainty and differing attacker and defender perspectives, the framework can still provide meaningful risk assessments even when information is noisy, missing, or conflicting. This allows decision-makers to focus defensive efforts where they matter most.
The second part, Intrusion Response Systems, develops an intelligent defense mechanism for vehicle networks, such as those used in autonomous or connected cars. Using deep reinforcement learning, the system learns how to respond to attacks in real time while accounting for uncertainty in attack detection and system behavior. The design reduces the success rate of attacks and improves safety-related outcomes like route completion and compliance with driving rules. A human-in-the-loop extension allows expert feedback to guide learning, making the system's behavior more interpretable and trustworthy in safety-critical settings.
The third part, Anytime Inference, focuses on decision-making under limited resources. In real-world systems, there is often not enough time or computing power to perform perfect analysis. The proposed Anytime Inference approach allows the system to produce progressively better answers as more resources become available and to stop early when needed while still providing useful results. This makes mission assessment and defense practical in fast-moving or constrained environments.
Overall, this dissertation shows that three elements are essential for resilient autonomous systems: explicitly accounting for uncertainty, anticipating strategic behavior by attackers and defenders, and adapting reasoning to available resources. By combining these ideas into a unified framework, the work advances the reliability, safety, and trustworthiness of autonomous systems operating in complex and adversarial environments. Future work can extend this approach to multi-agent systems, deeper human collaboration, and even larger and more complex real-world applications
PeerJ
No full textMeeting the needs of dogs in a typical animal shelter can be a challenging proposition. Negative environmental inputs, such as excessive noise, restrictive kenneling, and social isolation, contribute to the compromised welfare that dogs experience. Human-animal interaction, such as a temporary stay outside of the shelter in a caregiver’s home, has been shown to reduce dogs’ cortisol levels and increase their rest. What is less understood is if longer durations of foster care could extend those benefits. In addition, dogs living with a conspecific in the shelter, co-housing, has been even less explored, but available findings suggest that dogs’ behavior can be improved by living with another dog. In the present study, we investigated the impacts of weeklong fostering on dogs’ urinary cortisol and activity. Two animal shelters, one open and one managed admission, participated. Exclusively at the open admission facility, a smaller sub-study explored the effects of co-housing prior to foster care (i.e., with and without a dog) and following (i.e., without another dog or with a familiar or new dog) in the animal shelter. To answer these research questions, dogs’ urine was collected in the morning for cortisol: creatinine analysis and activity monitors were worn by the dogs for 17 days: five days in the animal shelter, seven days in a caregiver’s home, and five days in the shelter following foster care. In total, 84 dogs participated with 1,385 cortisol:creatinine values and 1,205 activity totals across five activity level types. At both shelters, we found dogs’ cortisol levels decreased, and they spent more time resting during weeklong fostering. Moreover, no significant differences in cortisol or activity were found pre- and post-fostering, with the exception of more time being spent in mid-intensity activity in the shelter following foster care as compared to before. These findings align with investigations of shorter durations of foster care, although the magnitude of the present intervention’s impact was greater. With regards to the type of housing dogs experienced (with or without another dog), no difference was found in dogs’ cortisol values in either the days before or after foster care with no effect on their activity detected pre-fostering; however, dogs’ activity was influenced by living with a familiar dog upon reentry to the animal shelter following foster care. Specifically, dogs rested more and engaged in less high activity, indicating a positive effect on their welfare. Lastly as has been previously observed, significant differences in cortisol and activity were found between our shelters, suggesting that environmental differences are contributing to canine welfare that require further scientific exploration. In total, a weeklong reprieve from the animal shelter, as well as co-housing with a familiar dog upon return to the shelter are two evidence-based interventions that can improve the welfare of shelter-living dogs.Published versio
Cell-specific network-based cell type prediction via graph convolutional network using transcriptomics profiles
No full textIdentifying cell types is crucial for characterizing biological phenomena in tissues at the single-cell level and understanding intracellular and intercellular interactions. Recent studies have introduced computational tools for cell type prediction using machine learning (ML), tailored for single-cell and spatial transcriptomics datasets. However, these approaches primarily focus on leveraging the gene expression profiles of individual cells, often overlooking the interactions between neighboring cells. Such interactions are vital, as they activate signaling pathways and coordinate gene expression. In this study, we introduce CSNpred, a cell type prediction framework that integrates graph convolutional networks with cellspecific network construction for transcriptomics data. Our model identifies neighboring cells with similar gene expression patterns, particularly those within close spatial proximity (when applicable) and constructs a network for each cell. This approach enables the learning of graph embeddings that account for both the cell’s gene expression and that of its neighbors. CSNpred outperforms the state-of-the-art cell type identification method and widely used ML-based classifiers, demonstrating superior prediction performance across various scenarios. Furthermore, we examined the role of cell-specific network construction in enhancing the classifier’s robustness, further validating its efficacy. CSNpred is publicly available at https://github.com/cbi-bioinfo/CSNpred.Published versio