Graph Network-Based Predictive Modeling for Next-Generation Smart Wearable Textiles

2.1 Smart textiles in application domains

Smart wearable textiles are increasingly deployed in healthcare, where continuous monitoring of electrocardiogram, respiration, and body movement supports personalized medicine and remote health management [12, 37, 40]. In sports and fitness, sensor-integrated fabrics provide real-time feedback for performance optimization [3, 8]. In defense and security, soldiers’ physiological status and environmental exposure are monitored through e-textiles, enhancing safety and operational effectiveness [5, 41]. Consumer markets have also seen the growth of interactive fashion and IoT-connected garments, embedding sensors for gesture recognition and augmented reality experiences [35, 42, 43].

2.2 Predictive modeling approaches in textiles

Traditional predictive modeling techniques in textiles have primarily concentrated on forecasting the mechanical or functional performance of yarns and fabrics. Methods such as regression analysis, support vector machines, and artificial neural networks have been used to predict tensile strength, comfort, or durability [38, 44]. In wearable systems, more advanced deep learning approach models have been applied to tasks such as analyzing physiological signals for health monitoring and human–computer interaction [19, 45]. While these models demonstrate strong performance in pattern recognition and sequential data analysis, they are limited by their assumption that data points are either independent or strictly temporal. This overlooks the interconnected and relational nature of innovation and technology evolution in smart textiles.

To address this gap, graph-based predictive modeling has emerged as a promising approach. Unlike conventional models, graph networks represent technologies, patents, or sensors as nodes, with their interactions (e.g., citations, classifications, or functional integration) modeled as edges. This enables the detection of clusters, hubs, and pathways that can signal emerging innovation domains. For example, in patent networks, centrality measures such as degree, betweenness, and eigenvector help identify influential technologies and their convergence potential, providing early signals of future trends.

2.3 Graph network analysis

Graph network analysis has recently emerged as a powerful technique to study complex, interconnected systems. In the context of technological innovation, patents can be represented as nodes, and their relationships, such as citations, co-classifications, or co-inventorship, as edges [17]. This representation enables the identification of hubs, clusters, and bridging technologies, offering deeper insights than frequency-based methods [46]. Applications in other domains, such as biotechnology and information technology, have demonstrated the potential of network analysis to uncover innovation pathways and predict emerging technologies [47, 48].

However, in the textile and wearable domains, the application of graph network analysis remains limited. Existing studies often apply simple bibliometric mapping or co-word analysis without fully leveraging structural network measures such as degree, betweenness, and eigenvector centrality. These omissions hinder the ability to detect influential nodes that act as gateways for technology convergence. Moreover, very few studies have combined network-based approaches with predictive modeling to forecast the evolution of wearable textile technologies. This creates a clear research gap: the need for a framework that applies graph network analysis to patent data in order to identify emerging hubs, predict convergence pathways, and guide R&D strategies in next-generation smart wearable textiles.

3.1 Data collection and preprocessing

Patent data were selected as the primary source due to its richness in technical disclosures, citations, classifications, and innovation trajectories [49, 50]. Patents represent one of the most comprehensive repositories of technological knowledge, as they not only disclose novel inventions but also provide structured information about the trajectory of innovation within a field [10]. Unlike general scientific publications, patents are closely tied to practical applications and commercialization efforts, making them particularly suitable for forecasting technological evolution. Data were collected from the global patent database USPTO, focusing on patents related to smart textiles by using the query “smart textile” OR “wearable textile” OR “e-textile” OR “conductive fiber” OR “biosensor fabric” OR “energy harvesting fabric.” Each patent record includes (1) bibliographic information (title, abstract, assignee, inventors, year), (2) citation network (forward and backward citations), (3) International Patent Classification (IPC) or Cooperative Patent Classification (CPC) codes for technology categorization, and (4) full text (claims and abstract).

Before constructing the graph network, the dataset underwent a preprocessing phase to ensure accuracy and consistency. Duplicate entries, incomplete metadata, and irrelevant patents were filtered out. Citation relationships were standardized to maintain directionality (from citing to cited patents), ensuring that the resulting network accurately represented the knowledge flow within the domain. Assignee names were also normalized to reduce redundancy, as the same organization often appeared under multiple variations.

To classify patents into distinct technological areas, the IPC system was used as the primary taxonomy. Each patent may be assigned multiple IPC codes, reflecting its interdisciplinary nature. For this study, we mapped patents to their primary IPC subclass to identify dominant technological domains, while retaining secondary codes for analyzing cross-domain interactions. For instance, patents categorized under D04H1/00 (woven or non-woven textiles) but also cross-listed with H05K7/02 (printed circuits or conductive layers) indicate a strong linkage between textile engineering and electronic integration. By leveraging IPC codes in this way, the graph network could be structured to highlight clusters of related technologies and detect areas where distinct technological fields such as materials science, electronics, and healthcare intersect.

3.2 Graph construction and network analysis

A patent network was constructed where nodes represent patents, inventors, or assignees, depending on the analytical perspective adopted. In the patent-centered view, each node corresponds to a single patent document, while in the alternative projections, nodes may represent inventors (to study collaboration structures) or assignees (to capture organizational-level interactions). Edges represent meaningful relationships between these entities. For patent-level graphs, edges can encode citation relationships, where a directed edge from patent i to patent j indicates that cites j. In co-classification networks, edges connect two patents that share at least one IPC or CPC code. In co-inventorship networks, undirected edges connect inventors who have jointly contributed to at least one patent. These different edge definitions allow for the exploration of knowledge flows, technological proximities, and collaboration dynamics within the innovation ecosystem. Formally, the adjacency matrix AA for the citation graph is defined as follows:

Here, A is generally sparse and asymmetric, as citation links are directed and relatively rare compared to the total number of potential connections. In contrast, for co-inventorship or co-classification networks, the adjacency matrix may be symmetric, reflecting the mutual nature of collaborations or common classifications. Weighted variants of A can also be constructed, for instance by assigning weights equal to the number of shared classifications or the frequency of co-authorship between inventors. Figure 2 illustrates the graph network generated from the IPC co-classification adjacency matrix.

Once the graph is constructed, standard network analysis techniques are applied. Centrality measures, such as degree, betweenness, and eigenvector centrality, help identify influential patents, key inventors, or dominant organizations. Community detection algorithms reveal clusters of patents that form technological domains or groups of inventors that collaborate closely. Structural properties, such as density, average path length, and clustering coefficient, provide insight into the overall topology of the network, indicating whether knowledge flows are centralized, decentralized, or fragmented.

4.1 Data exploration

The dataset collected from the USPTO through wispon.com contained 1,158 patents related to wearable and smart textiles between 2002 and 2021. The temporal distribution in Figure 3 illustrates a steady increase in filings after 2015, coinciding with the rapid commercialization of fitness trackers and the growth of IoT applications in textiles. The trend suggests that innovation in this field is not only persistent but also accelerating, particularly in recent years, where sustainability and health monitoring have become dominant themes.

The classification of patents by IPC codes highlights the technological diversity of wearable textiles. Figure 4 shows the top 15 IPC classifications for wearable textile patents, where A61B (medical diagnosis and surgery) dominates, reflecting the strong role of health and biomedical applications. This is followed by G06F (digital data processing) and A41D (garments), showing the integration of computing technologies into clothing design. Other categories, such as D06F (textile care), H05K (printed circuits), G01N/G01L (sensing and measuring), and A61F/A61M (medical devices), indicate the growing convergence of textiles with sensors, electronics, and healthcare solutions. Meanwhile, smaller groups such as G06K, G08B, B65H, B32B, and G06Q highlight cross-domain opportunities in smart apparel, layered materials, and IoT-based data services. Overall, the IPC distribution emphasizes that wearable e-textile innovation emerges from the intersection of healthcare, digital processing, and textile engineering, pointing to rich opportunities for future convergence technologies.

To provide a more detailed view, Table 1 maps the top IPC codes to their respective technology areas and illustrates their role in the smart wearable ecosystem.

The distribution of the number of claims across the top 15 IPC codes in wearable textile patents provides insights into the complexity and scope of technological protection within different domains. As shown in the boxplot in Figure 5, categories such as D03D (woven fabrics) and B32B (layered products) exhibit wider variability and higher median claim counts, indicating broader and more complex inventions that often encompass multiple technical aspects. In contrast, classes such as A61B (medical diagnosis, surgery, identification) and G06F (digital computing) show a more concentrated range of claims, suggesting more standardized innovations with narrower scopes of protection. Outliers in several IPC codes, such as A61F and G06Q, highlight individual patents with unusually high claim numbers, reflecting attempts to secure extensive protection in emerging areas such as smart medical wearables and computational systems for textiles. Overall, the variation in claim distributions emphasizes the heterogeneity of innovation strategies across wearable textile technologies, where some domains focus on incremental, specialized patents, while others pursue broader, multifunctional protections.

4.2 Network analysis of innovation pathways

To identify emerging technological opportunities, the patent dataset was represented as a graph network, where nodes represent patents and edges denote relationships such as citations or shared IPC classifications. Network analysis was then performed to measure structural indicators of technological importance. The next step in the analysis involves constructing a network representation of IPC code co-occurrences to uncover technological linkages within wearable textile patents. Using the Python code above, a co-occurrence matrix (adjacency matrix) was generated, where each node represents an IPC code and each edge reflects the frequency with which two codes appear together in the same patent. This approach captures the multidimensional nature of innovation, as patents often span multiple technical domains. The process begins by extracting all IPC codes from the dataset, identifying unique codes, and initializing a square adjacency matrix with zeros. For each patent, combinations of IPC code pairs were computed, and the matrix was incremented to record the number of times two codes co-occur. The resulting adjacency matrix provides the foundation for building a graph network that represents the interconnections between different technological areas. This network enables further analysis of structural properties such as degree centrality, betweenness centrality, and clustering coefficients, which help identify highly connected technologies, bridging domains, and clusters of emerging innovations. By examining these relationships, the study highlights not only the dominant technological areas but also the convergence points where novel opportunities for smart wearable textiles are likely to emerge.

To complement the adjacency matrix, the co-occurrence relationships were further visualized as a graph network, as shown in Figure 6, where nodes represent IPC codes and edges indicate their frequency of co-appearance in patents. The thickness of the edges corresponds to the strength of the relationship (i.e., the number of co-occurrences), while node size is scaled according to degree centrality, reflecting the relative importance of each technological area within the network. This visualization provides an intuitive map of the technological landscape, allowing clusters of closely related IPC codes to emerge clearly. For example, medical-related classes such as A61B and A61F are shown to form strong connections with computational categories such as G06F, highlighting the convergence of healthcare and digital technologies in wearable textiles. By presenting the network graph, stakeholders can quickly identify highly connected domains and innovation hot spots, making it a valuable tool for forecasting technological trajectories and guiding R&D investment in next-generation smart e-textiles.

The centrality analysis highlights the most influential IPC codes within the wearable textile patent network in Table 2, providing insights into the technological domains most likely to drive future convergence and innovation opportunities. Across multiple centrality measures, A41D (garments and protective clothing), G06F (digital computing), B32B (layered products), and A61B (medical diagnosis, surgery, identification) consistently appear as top-ranked nodes. This dominance suggests that the integration of advanced fabrics with digital technologies and medical applications will remain central to next-generation wearable textile innovation. For instance, A41D’s high values in degree, betweenness, and eigenvector centrality indicate its pivotal role in connecting clothing with other technological fields, positioning it as a hub for convergence in smart apparel and protective textiles. Similarly, G06F and A61B underscore the growing overlap between computation, sensing, and healthcare applications, pointing toward opportunities in AI-driven wearables and digital health monitoring. The prominence of B32B reflects the strategic importance of multilayered and composite textile structures, which enable energy harvesting, embedded sensors, and enhanced durability in e-textiles. Supporting domains such as D03D (woven fabrics), H05K (printed circuits), and G01N (material analysis and sensing) also appear strongly, indicating that the convergence of traditional textile engineering with electronics and sensor technologies is a critical enabler. Overall, the centrality results suggest that future breakthroughs will likely emerge at the intersections of garment design, digital computing, medical devices, and layered textile composites, providing clear pathways for innovation in healthcare, sports, and IoT-driven applications of smart wearable textiles.

Table 3 shows that future wearable textiles will evolve through convergence between traditional textile engineering (A41D, D03D, D02G) and advanced domains such as digital computing (G06F), medical technologies (A61B, A61F), sensing (G01N), and electronics (H05K). Healthcare and medical applications are projected to be the strongest drivers, leveraging biocompatible fabrics and AI-based diagnoses. Energy sustainability via self-powered e-textiles will be crucial for IoT integration. Eco-friendly and recyclable solutions will align with global sustainability agendas.

4.3 Linking wearable textile trends to Industry 4.0/5.0

The development of smart wearable textiles does not occur in isolation but is tightly connected to the broader paradigms of Industry 4.0 and the emerging Industry 5.0. While Industry 4.0 emphasizes digitalization, connectivity, and real-time data analytics [51], Industry 5.0 expands this vision by integrating human-centric design, ergonomics, and sustainability [52, 53]. Wearable e-textiles embody these principles by serving as both enablers and beneficiaries of industrial digital transformation. Table 4 presents a compact mapping of headline trends in wearable textiles against Industry 4.0/5.0 functions, industrial touchpoints, and key performance indicators (KPIs).