Forthcoming Articles
International Journal of Information and Communication Technology
Forthcoming articles have been peer-reviewed and accepted for publication but are pending final changes, are not yet published and may not appear here in their final order of publication until they are assigned to issues. Therefore, the content conforms to our standards but the presentation (e.g. typesetting and proof-reading) is not necessarily up to the Inderscience standard. Additionally, titles, authors, abstracts and keywords may change before publication. Articles will not be published until the final proofs are validated by their authors.
Forthcoming articles must be purchased for the purposes of research, teaching and private study only. These articles can be cited using the expression "in press". For example: Smith, J. (in press). Article Title. Journal Title.
Articles marked with this shopping trolley icon are available for purchase - click on the icon to send an email request to purchase.
Online First articles are also listed here. Online First articles are fully citeable, complete with a DOI. They can be cited, read, and downloaded. Online First articles are published as Open Access (OA) articles to make the latest research available as early as possible.
Articles marked with this Open Access icon are Online First articles. They are freely available and openly accessible to all without any restriction except the ones stated in their respective CC licenses.Register for our alerting service, which notifies you by email when new issues are published online.
International Journal of Information and Communication Technology (25 papers in press) Regular Issues
Abstract: In order to explore strategies for enhancing the information technology education capabilities of university teachers, this study proposal is an overall model based on the knowledge framework of technical teaching content, and improvements. Experimental results show that the PA dimension scored the highest (M = 3.95) and the TPCA dimension scored the lowest (M = 1.70), and other dimensions scored between 2.80 and 3.50. A comparison before and after the implementation of the promotion strategy revealed the most significant improvement among teachers aged 41 to 50, with an increase of 53.90%. The proportion of teachers over 50 years old also increased from 8.89% to 40.63%. The strategies proposed in this study have reference value for research on improving various aspects of teacher capabilities, can promote teacher development, and play a role in driving educational industry reform. Keywords: TPACK; informatisation; education capacity; APNN; DDAE-SVR; BPNN. DOI: 10.1504/IJICT.2026.10076854
Abstract: With the growing demand for precise cross-dialect syntactic analysis, this work proposes an end-to-end framework that automatically labels English syntactic variants and quantifies their distribution across dialects. The approach integrates parser-generated silver annotations, a human-audited gold subset, and a dual-head neural model combining CRF-based sequence tagging and span classification. Domain adaptation with gradient reversal, moment matching, and supervised contrastive learning enhances robustness to dialectal shift, while probability calibration ensures accurate rate estimation. Evaluations on multi-source corpora covering American, British, Australian, and Indian English show that the proposed model improves out-of-dialect macro-F1 by 6.9 points over a strong RoBERTa baseline, reduces domain divergence in encoder space by over 55%, and recovers stable, interpretable contrasts for variants such as that-complementiser drop, particle movement, and dative alternation. Keywords: syntactic variation; automatic labelling; cross-dialect analysis; CRF; contrastive learning. DOI: 10.1504/IJICT.2026.10076855
Abstract: To analyse emotions and predict dissemination trends of Sichuan handicraft intangible cultural heritage (ICH) inheritors, this study proposes an ALBERT-based sentiment analysis model and a TCN-based prediction model. Integrating multi-dimensional feature extraction, attention mechanisms, and a dual-channel TCN structure, the models capture long-term dependencies by fusing emotional features with dissemination indicators. Experiments on social media data show the improved ALBERT model achieves 86.21% precision and 83.97% recall in sentiment analysis, outperforming BERT, RoBERTa, and LSTM while reducing memory usage. The TCN-based prediction model attains MAE of 0.389%, MSE of 0.007%, and 91.36% fit, improving accuracy and stability. Emotional distributions exhibit periodic fluctuations during festivals, with dissemination trends driven by both emotional and behavioural indicators. This research enhances ICH emotion analysis and prediction technologies, revealing group emotion patterns and communication evolution to provide data support for government short-video timing decisions and inheritors' content optimisation. Keywords: sentiment analysis; ALBERT; attention mechanism; temporal convolution network; TCN; dissemination trend. DOI: 10.1504/IJICT.2026.10076856
Abstract: Amid globalisation and growing cross-language information needs, machine translation is crucial for overcoming language barriers. Deep learning has advanced it, but transformer faces limitations: insufficient efficiency in capturing long-range dependencies and poor performance in low-resource translation. To address these, this study proposes three core solutions: 1) a hybrid LSTM-transformer architecture fusing LSTMs gating mechanism (long-sequence modelling) and transformers self-attention (global context capture); 2) an adaptive gradient clipping (AGC) strategy for training stability; 3) dynamic weight sharing with adversarial domain adaptation to enhance cross-language transfer. Experiments on WMT14 English-German/French corpora show the models BLEU value is 2.8 higher than benchmark Transformer, with 18% faster convergence; in English Romanian low-resource scenarios, the transfer mechanism boosts BLEU by 5.3. This study validates the hybrid architecture and optimisation strategies, offering new ideas for efficient gradient optimisation and low-resource translation models. Keywords: machine translation; LSTM-transformer; gradient cropping; cross-language transfer; adversarial learning. DOI: 10.1504/IJICT.2026.10076857
Abstract: This study focuses on emergency scheduling for a photovoltaic-fuel cell hybrid microgrid under fault conditions, and proposes an online optimisation strategy that integrates model predictive control (MPC) and mixed-integer linear programming (MILP). The method coordinates the management of fuel cell output, energy storage charging/discharging constraints, photovoltaic power curtailment, and interruptible load shedding within a rolling time horizon. It also incorporates the fuel cell health cost into the objective function to mitigate the stress caused by frequent start-stop operations and rapid power ramping. Simulations based on MATLAB/ YALMIP/Gurobi demonstrate that the proposed strategy can restore power balance and ensure zero interruption for critical loads within 20 seconds under typical scenarios such as photovoltaic power drop, load surge, single machine outage, and multi-fault compound disturbances. Compared to the baseline droop strategy, the approach significantly reduces photovoltaic curtailment, maintains the state of charge (SOC) within a safe range, and substantially decreases health costs and power fluctuations (30-49%) reduction in representative scenarios), validating the efficiency and robustness of the method under complex disturbances. Keywords: photovoltaic power generation; fuel cell; fault scenarios; emergency optimisation scheduling. DOI: 10.1504/IJICT.2026.10076950
Abstract: This paper proposes a model for English-derived place name recognition and translation using a knowledge graph and a phonetic generation algorithm. By integrating multiple algorithms, the model enhances both recognition and translation accuracy. Experimental results show an AUC of 0.892. With 100 training iterations, the recognition error rate is 1.3%, the translation error rate is 0.8%, and the BLEU score reaches 67.3%, demonstrating strong performance. Practical analysis indicates the model has the lowest time consumption, minimal memory usage, superior classification performance, and over 95% fluency and consistency. The innovation of the research lies in the construction of a bidirectional dynamic interaction fusion mechanism through a knowledge graph, an LSTM algorithm, and a bidirectional matching maximum algorithm, targeting the semantic specificity of English-derived place names. This breaks the traditional one-way static fusion and achieves precise scene-based collaboration and closed-loop optimisation of semantics and phonetics. Keywords: knowledge graph; phonetic generation algorithm; english-derived place names; recognition and translation; LSTM; bidirectional maximum matching algorithm. DOI: 10.1504/IJICT.2026.10076951
Abstract: To address issues such as fuzzy topological structures, overlooked group differences, and the disconnect between visualisation and quantitative analysis in course grade-teaching evaluation correlation studies, this study proposes an integrated model based on Fruchterman-Reingold and Theil-Sen. Its core innovation lies in constructing a dual-module collaborative architecture: enhancing course community identification through spectrum-guided layout optimisation, and employing topology-feature weighted group regression that integrates topological stability indices with subgroup trend medians to precisely characterise heterogeneous group associations. It implements a closed-loop analytical paradigm of topological feature extraction group difference modelling feedback optimisation, overcoming the limitations of linear processes that separate network layout from regression validation. Experimental results demonstrate a convergence efficiency of 0.77%/iteration, outlier robustness of 0.89, and processing time of 87.2 ms. The model achieved a correlation estimation bias of 0.10, group difference identification accuracy of 0.94, and cross-discipline generalisation error of 0.10. In loosely structured course groups, performance declined notably. This model significantly enhances the analytical capability for curriculum interrelationships and improves the accuracy of cross-group correlation estimation in educational assessment, providing reliable technical support for dynamic monitoring of teaching quality and optimisation of interdisciplinary curriculum systems. Keywords: Fruchterman-Reingold; Theil-Sen; course grades; evaluation grades; educational data mining; topological analysis; robust regression. DOI: 10.1504/IJICT.2026.10076952
Abstract: This study proposes a Transformer-based cross-cultural intelligent translation system to enhance international communication. By integrating attention mechanisms and large-scale multilingual datasets encompassing 47,850 samples across seven languages from 35 countries, the model achieves 97.3% accuracy in predicting language competency while ensuring contextual fluency and cultural adaptability. The approach outperforms traditional and BERT-based methods, offering a scalable solution for multilingual, multicultural contexts. Language is a vital bridge for cross-cultural communication, especially in global collaborations. However, traditional translation systems struggle with contextual accuracy and cultural inclusivity. Previous studies have explored neural machine translation enhancements, such as GANs, BiLSTM generators, and syntax-aware methods. While effective, these approaches often face limitations in low-resource languages and cultural adaptability. A hybrid deep learning framework combining Transformer architecture and attention mechanisms was developed. The proposed model achieved 97.3% accuracy, 96.8% precision, 95.6% recall, and 96.4% F1-score. These results outperform state-of-the-art baselines, demonstrating superior performance in cross-cultural translation. Keywords: transformer model; cross-cultural communication; neural machine translation; NMT; natural language processing; NLP; attention mechanism; legislative impact assessment; LIA. DOI: 10.1504/IJICT.2026.10076953
Abstract: This study develops a framework for urban spatial morphology evolution and resilience assessment through multi-source data fusion. Focusing on Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Pearl River Delta (PRD) from 2010 to 2023, the research integrates satellite imagery, social sensing data, and socioeconomic statistics using hierarchical deep learning. The methodology achieves 92.4% classification accuracy, a 24.1% improvement over single-source methods. Analysis reveals distinct evolution patterns: radial expansion in BTH (4.2% annually), polycentric coalescence in YRD (5.8%), and corridor-oriented development in PRD (6.1%). Resilience assessment demonstrates YRDs superior performance (0.86 index), capturing COVID-19 impacts with 2%-3% decline followed by rapid recovery. Contributions include integrating complex adaptive system theory with spatiotemporal fusion and developing hierarchical architectures with dynamic resilience models. Keywords: urban morphology evolution; multi-source data fusion; urban resilience assessment; deep learning; complex adaptive systems; spatiotemporal analysis. DOI: 10.1504/IJICT.2026.10076954
Abstract: To better control internal financial risk in industrial enterprises, this study developed an intelligent financial internal control system based on robot process automation. The system utilises a multi-objective evolutionary clustering algorithm with reference vectors to enhance credit management and billing optimisation. Results demonstrate an average value of 0.95 for the proposed method. The maximum mean of the normalised mutual information entropy index is 0.92 under the normalised mutual information index. This algorithm exhibits high clustering accuracy and robustness. The system achieves an accuracy rate of over 0.8 for enterprise credit rating, with an average of 86.9%. It surpasses other systems in evaluation accuracy and exhibits consistent results in manual evaluation. It can be seen that the performance of the system is more excellent, which can help industrial enterprises to carry out internal control of financial risks and simplify the financial workflow. Keywords: sustainable; financial management; robotic process automation; industrial enterprises; financial internal control. DOI: 10.1504/IJICT.2026.10076955
Abstract: This study presents a WebGPU-based visualisation framework aimed at enhancing the processing, rendering, and interactivity of large-scale, multidimensional datasets in educational contexts. By integrating artificial intelligence (AI) tools with advanced visualisation techniques, the framework enables efficient data preprocessing, interpolation, and volume texture generation for seamless web-based visualisation. Utilising datasets from oceanographic simulations and educational performance metrics, the system demonstrates versatility across domains. Comparative experiments show that the WebGPU- based solution significantly outperforms previous WebGL-based implementations, reducing rendering time and increasing frame rates. User surveys report high satisfaction in functionality, personalisation, usability, and compatibility. These findings highlight the potential of AI-driven educational data analytics and visualisation tools to support decision-making, enhance user engagement, and promote data literacy in academic and professional training environments. Keywords: big data visualisation; web GPU; data analysis teaching platform; volume rendering; artificial intelligence in education; educational data analytics; visualisation framework; user satisfaction; data preprocessing; interactive visualisation. DOI: 10.1504/IJICT.2026.10076979
Abstract: Multimodal sentiment analysis remains challenging due to the difficulty of fusing heterogeneous data like facial expressions, speech, and pose. Unlike unimodal analysis, it is often hindered by problems of information redundancy, heterogeneity, and complex temporal dynamics. This paper proposes MERCAT, a Transformer-based model featuring a cross-modal self-attention mechanism to capture deep correlations between different modalities. This design enables highly efficient and context-aware inter-modal fusion. Extensive experiments on multiple benchmarks show that MERCAT achieves excellent performance. It notably excels in emotion classification, significantly improving accuracy and F1-score over strong baselines, and in emotion intensity prediction, where it substantially reduces error and improves correlation. The study conclusively verifies the efficacy of the cross-modal attention mechanism for information fusion, providing a robust and effective solution for advancing multimodal sentiment analysis. Keywords: multimodal sentiment analysis; sentiment feature extraction; information fusion; transformer architecture; cross-modal self-attention mechanism; sentiment intensity prediction. DOI: 10.1504/IJICT.2026.10077026
Abstract: Fatigue monitoring is essential for athletic training. This study addresses class imbalance and scarcity of severe fatigue samples by proposing a spatio-temporal graph convolutional network enhanced with a generative adversarial network. A conditional Wasserstein generative adversarial network generates realistic synthetic skeletal sequences to expand the training set. Combined with spatio-temporal feature extraction, our model achieves end-to-end fatigue level classification. Evaluations on the daily life activities dataset demonstrate superior performance, with accuracy, precision, recall, and F1-score reaching 92.5%, 91.8%, 92.2%, and 92.0% respectively outperforming support vector machine by 20.2%, long short-term memory by 11.0%, baseline spatio-temporal graph convolutional network by 3.8%, and variational autoencoder-augmented models by 2.8% in accuracy. Ablation studies validate both the generative adversarial network augmentation and nonlinear labelling strategy, offering a reliable vision-based framework for fatigue monitoring. Keywords: generative adversarial networks; GANs; spatio-temporal graph convolution; exercise fatigue monitoring; data enhancement; Wasserstein generative adversarial networks. DOI: 10.1504/IJICT.2026.10077027
Abstract: This paper introduces a lightweight enhanced model based on YOLO11 to tackle the dual challenges of reducing model size and boosting precision for deploying traffic sign detection systems on autonomous driving mobile devices. Initially, the backbone network integrates the C3G2 module, which effectively minimises parameters while strengthening multi-scale feature extraction. Subsequently, a parameter-free attention mechanism, Tr-simAM, is developed to lower computational load and heighten sensitivity to small targets. Finally, the neck network incorporates the lightweight dynamic upsampling module DySample to enhance feature fusion and localisation precision. Experimental outcomes reveal that the refined algorithm achieves an 84% precision rate, signifying a 1.2% gain over the baseline YOLOv11n model, while model parameters and computational resources decrease by 19% and 7%, respectively. These lightweight enhancements efficiently fulfil practical requirements for real-time performance and computational constraints in autonomous driving mobile devices. Keywords: YOLO11; traffic signs; object detection; lightweight network; attention mechanism; upsampling module; private dataset. DOI: 10.1504/IJICT.2026.10077028
Abstract: With the acceleration of global population aging and the rising proportion of Chinas elderly population, the aging design of residential space is of great significance to improve the quality of life of the elderly. From a biological perspective, elderly people have special needs for living space due to physical decline, sensory degradation, and cognitive changes, while traditional living spaces have obvious deficiencies in spatial scale, functional zoning, and environmental adaptation. This study is based on the biological characteristics of elderly people, deeply integrating computer and software engineering technologies, and proposing optimisation strategies from three dimensions: spatial layout, furniture and facility configuration, and environmental creation to address the existing problems in traditional residential spaces. The research aims to improve the convenience of daily activities and indoor environment comfort for the elderly, effectively enhancing their living safety, convenience, and comfort. Keywords: residential space; suitable for aging; computer and software engineering; spatial layout; biological characteristics. DOI: 10.1504/IJICT.2026.10077029
Abstract: Vast content on social media offers a unique perspective for understanding academic stress. However, the multimodal nature of social media data, coupled with its high dimensionality and complexity, poses significant challenges for quantifying perceptions of academic stress. To address this, this paper first optimises the knowledge distillation algorithm based on gated networks. Then, with the text modality as the core, it employs a cross-self-attention mechanism to achieve deep integration of social text, visual, and audio modalities. The fused information serves as the teacher model, while the audio and visual modalities act as student models. The multimodal knowledge distillation module transfers academic stress sentiment information from the text modality to the other modalities, enhancing the models perception capabilities. Experimental results demonstrate that the proposed model reduces mean absolute error by at least 25.9%, enabling more precise quantification of users academic stress levels from social media data. Keywords: social media; academic stress identification; knowledge distillation; multimodal features; attention mechanism. DOI: 10.1504/IJICT.2026.10077124
Abstract: The rapid development of intelligent communication technologies has brought innovative opportunities to the field of music education, as traditional models are no longer sufficient to meet students learning needs. This study employs a convolutional neural network (CNN) as the core algorithm to process audio and dance data, extracting high-level features such as rhythm and body movements through its multi-layered hierarchical structure. A support vector machine (SVM) algorithm is used to assess student abilities, while the convolutional neural network processes data to extract features, and intelligent sensor technology is integrated to build a teaching platform. The study found that the support vector machine achieved an accuracy of 93.7% in music feature classification, while the convolutional neural network improved the accuracy of dance movement classification to 96.3%. This model significantly improves the accuracy of teaching assessment, providing an intelligent solution for music and dance education and promoting human computer interactive teaching. Keywords: intelligent communication technology; constitutional neural network; CNN; music and dance teaching; intelligent sensor; artificial intelligence. DOI: 10.1504/IJICT.2026.10077125
Abstract: In this paper, we deal with the shortcomings of the old brand sentiment watching ways which are unable to catch those deep cause-effect links as well as their ever-changing development trends by presenting a new kind of brand sentiment watching way a dynamic growth watching system for brand sentiment using causal finding neural networks. This model can find the causal structure between different sentiment elements by itself with a structured causal discovery module. Combined with a temporal neural network to model the sentiment evolution path, it can do causal inference and dynamic prediction of sentiment trend. From the experiments, we see that our method is much better than the other models on all the metrics, and the improvement is statistically significant. Keywords: causal discovery neural network; brand sentiment; dynamic forecasting. DOI: 10.1504/IJICT.2026.10077126
Abstract: This paper presents a professional competence indicator system for undergraduate architecture students. Grounded in national policy documents, theories and international frameworks, the system covers five dimensions with 51 indicators. Policy-text analysis, principal-component analysis and online recruitment data mining were integrated, with indicator weights assigned through a weighted-coefficient method and dynamically updated. A graduate-employer dual-evaluation model and a curriculum-reform pilot were introduced to verify feasibility and effectiveness. The system offers an operational instrument and reference framework for optimising architecture talent cultivation programmes and consolidating industry-education symbiosis. Keywords: industry-education integration; architecture; professional competence indicator system; educational reform; talent cultivation. DOI: 10.1504/IJICT.2026.10077141
Abstract: With the growing demand for personalised tourism, route recommendation has become a key issue in intelligent travel. Existing methods face limitations in personalisation, temporal rhythm, and adaptability to dynamic environments. We propose an interest-aware and context-adaptive route recommendation model (ICRR). First, an interest-aware adaptive attention mechanism integrates user interest vectors into graph attention networks to enable personalised representations. Second, a temporal segmentation optimiser leverages LSTM and attention to capture temporal dependencies and solve the orienteering problem with time constraints, using adaptive perturbation search to avoid local optima. Finally, a dynamic route refinement mechanism models environmental factors through reinforcement learning for real-time route adjustment. Experiments show that ICRR outperforms baselines in user satisfaction, recommendation accuracy, and robustness, offering an efficient solution for smart tourism and intelligent transportation. Keywords: interest-aware attention; route optimisation; temporal modelling; reinforcement learning; personalised recommendation. DOI: 10.1504/IJICT.2026.10077142
Abstract: Childrens drama is an important way to help kids learn to express themselves through language. The classic interactive story style has problems, though, like having fixed content and not being able to personalise it. This study proposes a generative adversarial network (GAN)-driven interactive narrative system for childrens drama. First, we combine data from multiple sources to get a better understanding of emotion and meaning. Next, we use GAN to dynamically create the plays content to make sure the story structure makes sense and fits the mood. Finally, we make sure the system is responsive, and the user experience is good by improving the way people interact with it. The experimental results indicate that the approach achieves scores of 0.87 and 0.82 in logical coherence and emotional accuracy, respectively. This substantially enhances childrens engagement and emotional resonance, demonstrating considerable practical usefulness and promotional potential. Keywords: generative adversarial network; GAN; children’s drama; interactive narrative; multimodal interaction. DOI: 10.1504/IJICT.2026.10077186
Abstract: Creative teams increasingly ask generative systems to deliver series of images that look like they belong together. To stop style drift across scenes and tiles, we propose art coherence reconstruction, a cross-modal, set-aware pipeline for artistic image generation. In our scheme, first we parse text and references to disentangle content from style; then we fuse statistics and descriptors into a compact style code backed by a small memory; finally we apply hierarchical modulation and set-level losses to steer colour, stroke, and grain consistently. On storyboards, character sheets, and mosaics, ArtCoRe lifts the coherence score from 0.56 to 0.71, halves drift from 0.22 to 0.10, raises alignment from 0.62 to 0.69, and cuts macro FID from 56.2 to 48.7. The result is stable palettes, cleaner seams, and predictable control for real projects. Keywords: AIGC; cross-modal understanding; style coherence; diffusion models; hierarchical modulation; style memory; set-level evaluation. DOI: 10.1504/IJICT.2026.10077187
Abstract: This study intends to build a hybrid deep learning model based on transformer and bidirectional long short-term memory network (BilSTM) using high-precision multi-source data of NBA players. First, this study identifies critical load periods and captures sequence global dependence using transformer encoders multi-head self-attention (MSA). Then, the BiLSTM layer is used to enhance the bidirectional time-series context modelling, and finally the scenario classification results are output. The experimental results demonstrate that in the load trend prediction task, the models mean squared error (MSE) decreases by 23.3% compared to traditional long short-term memory (LSTM), and by 13.8% compared to the pure transformer model. While the R2 increases to 0.944, with an improvement of approximately 1.7%, indicating superior fitting accuracy and stability. This model demonstrates significant advantages in both recognition accuracy and generalisation capability. Overall, the proposed method is more sensitive and stable in capturing exercise loads complex temporal dependencies. Keywords: training load analysis; transformer model; bidirectional long short-term memory network; time series data mining; competitive sports. DOI: 10.1504/IJICT.2026.10077188
Abstract: Intangible cultural heritage paintings often suffer from fading, structural damage, and detail loss due to environmental and material aging, making high-fidelity digital reconstruction essential for cultural preservation. To address the limitations of red-green-blue imaging (RGB) and insufficient spectral-spatial modelling, this study proposes context-aware multi-spectral imaging network (CA-MSI-Net), a reconstruction method integrating multi-spectral imaging (MSI) with an improved U-Net architecture. Spatial and channel transformer modules are embedded to enhance long-range spatial dependencies and cross-band spectral interactions, while contextual modelling and multi-scale attention mechanisms strengthen texture perception and boundary restoration. Experiments on multispectral datasets demonstrate that CA-MSI-Net achieves superior reconstruction accuracy, with mean intersection over union (mIoU), dice similarity coefficient (DSC), and F1 score reaching 81.3%, 92.1%, and 0.94, respectively, outperforming UCTransNet and dynamic optimisation vision network (DovNet). The method shows robust performance across painting styles, lighting conditions, and spectral configurations. Keywords: intangible cultural heritage painting; digital reconstruction; multi-scale attention mechanism; multi-spectral imaging; U-Net. DOI: 10.1504/IJICT.2026.10077189
Abstract: With the development of artificial intelligence and sensor fusion, student psychological stress classification is shifting from subjective assessment to intelligent recognition based on multi-source physiological signals. However, existing methods often rely on single-channel signals and insufficient temporal and multimodal modelling. This study proposes a stress classification method based on sensor data fusion and neural networks. Multi source EDA, PPG, and body temperature signals are synchronously collected, denoised using biased Kalman filtering, and jointly modelled by 1D-CNN and BiLSTM to capture local and global temporal features. Experimental results show that CNN and BiLSTM achieve accuracies of 78.32% and 82.15%, respectively, while the fused model reaches 89.47% with an F1-score of 0.88. Multimodal fusion improves the F1-score to 0.90, exceeding single-modal EDA (0.70) and PPG (0.71). Kalman filtering reduces RMSE by 42.5% and increases entropy by 2.0 bits, enhancing classification performance and providing technical support for psychological stress management and intervention. Keywords: sensor data fusion; neural network; classification of student psychological stress; biased Kalman filter; BiLSTM. DOI: 10.1504/IJICT.2026.10077190 |
Open Access
Keep up-to-date
-
Our Blog -
Join us on Bluesky -
Follow us on X -
Visit us on Facebook - Our Newsletter (subscribe for free)
- RSS Feeds
- New issue alerts
- Inderscience is a member of publishing organisations including:
-
-
