International Journal of Intelligent Information and Database Systems (19 papers in press)

Regular Issues

• Hybrid jellyfish search sine cosine optimisation-based deep learning for big data classification using MapReduce framework on epileptic seizure data  
  by Jamunadevi Chandrasekar, Arul Ponnusamy 
  Abstract: The increase in the amount of big data with the technical advances makes the traditional software tools face difficulties and unable to handle them. In the medical field, big data technologies require new frameworks to leverage them. This paper proposes a novel big data classification using a MapReduce framework on epileptic seizure data is proposed. Here, the big data classification is accomplished in a MapReduce framework, wherein the mapper phase is applied with the data partitioned by deep embedded clustering (DEC). The classification is carried out in the reducer phase, where a deep long short-term memory (DLSTM) trained using the jellyfish search sine cosine (JSCS) algorithm is used for epileptic seizure detection based on the salient features determined from the EEG data. The JSCS-DLSTM is investigated for its efficiency based on accuracy, specificity, and sensitivity and is found to record superior values of 0.915, 0.927, and 0.919, respectively.
  Keywords: deep embedded clustering; DEC; jellyfish search sine cosine algorithm; JSCS; deep long short-term memory; big data classification; epileptic seizure detection.
  
  
• Sentiment classification method of fresh agricultural product reviews based on semantic and emotional optimisation  
  by Yindong Dong, Yu Zhang, Guodong Wu, Xia Chen, Lijing Tu, Guohua Fan 
  Abstract: The complexity of semantics and structure in fresh agricultural product reviews can lead to sparsity in the distribution of text sentiment information. To address this issue, a sentiment classification model (Electra XLNet BIGRU multi-head attention - EXBMA) for fresh agricultural product reviews is proposed, which combines semantic and sentiment information optimisation. Firstly, the Electra model and XLNet model were used to obtain word level and sentence level information of fresh agricultural product reviews, respectively, and the semantic features obtained were fused using (multimodal compact bilinear - MCB) algorithm. Secondly, TextRank is used to extract keywords to construct an emotional key dictionary and combined with a multi-head attention mechanism to enhance emotional attention. Finally, to enhance the contextual representation, the BIGRU is used to learn contextual information to improve the classification performance. The experimental results indicate that the EXBMA can better achieve collaborative optimisation of semantic and emotional information, and performs better than other existing classification models in emotional classification of fresh agricultural product reviews.
  Keywords: fresh agricultural products; emotional classification; semantic optimisation; sentiment lexicon; attention mechanism.
  
  
• Hybrid brave-hunting optimisation for heart disease detection model with SVM coupled deep CNN  
  by Pravin M. Tambe, Manish Shrivastava 
  Abstract: This research proposes a novel hybrid optimisation method called brave-hunting optimisation (BHO) inspired by lion optimisation (LO) and coyote optimisation (CO). The BHO algorithm is employed to fine-tune the SVM parameters, enhancing its classification performance. Simultaneously, a deep CNN model extracts complex and informative features from medical data. The combined approach capitalises on the strengths of both optimisation techniques to create a robust and accurate model for heart disease diagnosis. The performance evaluation of our model is conducted using comprehensive metrics, which achieve an accuracy of 94.89%, an F1-score of 94.48%, a precision of 94.48%, and a recall of 94.58% for a 90 TP. In the context of a ten k-fold evaluation, achieved 94.78% accuracy, 94.36% F1-score, 94.55% precision, and 94.13% recall.
  Keywords: cardiovascular disease detection; support vector machines; deep convolutional neural network; DCNN; brave-hunting optimisation; BHO; early prediction.
  
  
• A comprehensive survey of text-based semantic similarity with potential applications  
  by Puneet Sharma, K. Yogeswara Rao Yogi, Kavitha Murugan, T. Sakthivel 
  Abstract: Text-based semantic similarity has recently become a popular research topic, and estimating the semantic similarity between entities for learning and decision-making is a challenging research problem. Semantic similarity quantitatively measures the informativeness between the data objects based on their properties and relationships. Due to the vast availability of measures developed by domain experts from different research fields, choosing an appropriate semantic similarity measure that suits a specific application context is challenging. The primary objective of this systematic literature survey is to offer a comprehensive view of semantic similarity and potential applications that emphasise the enormous diversity of research contributions in a wide range of fields. In this context, it examines various categories of semantic similarity based on the underlying principles of the state-of-the-art approaches with pros and cons. Notably, the prospective application areas are explored extensively by highlighting their significance. Finally, the survey concludes by summarising valuable future research directions.
  Keywords: semantic similarity; natural language processing; NLP; corpus-based; knowledge-based; semantic similarity applications; deep learning; DL.
  
  
• IoT networking for worker safety monitoring using the construction site images and worker health records  
  by Leena Rakesh Jadhav 
  Abstract: The ability to enhance the construction workers safety performance on-site may be made possible by computer vision-based approaches and deep learning developments. However, due to a variety of technical accuracy, reliability, and administrative issues, the practical application of computer vision and deep learning has been constrained. Hence, to address the above limitations in the conventional techniques, this research develops a latran timber optimisation-based ensemble classifier (ensemble-based-LTO) to construct an IoT-enabled safety framework for construction workers. Utilising the bandwise texture descriptor and pre-trained weights from various architectures such as Resnet-101 and VGG-16, the proposed method extracts the refined features from ROI minimising the computational complexity. The optimised ensemble classifier, effectively learns the long-term contextual dependencies and the spatial properties which in turns increases the detection accuracy. According to the experimental validation, accuracy attainment at 80% of training is 96.74%, sensitivity attainment is 95.33%, and specificity attainment is 97.68%.
  Keywords: BiLSTM; convolutional neural network; CNN; IoT-enabled safety framework; latran timber optimisation; grey wolf optimisation.
  
  
• Subway tunnel deformation monitoring based on 3D laser scanning technology  
  by Xiaoming Ji, Xu Wu, Yan Bao 
  Abstract: As urban infrastructure, particularly subway systems, gets more intricate, it becomes crucial to prioritise the upkeep of its structural stability and safety. To guarantee the safety, structural stability, and operational effectiveness of subway tunnels, the study proposes using the TunnelScanPro framework, which uses 3D laser scanning technology to monitor and analyse changes in the design and form of subway tunnels over time. This study uses a three-dimensional laser scanning technology to monitor changes in the geometries of subway tunnels. More specifically, the study focuses on using visual information to evaluate the structural integrity and safety of the tunnels. The paper aims to use 3D laser scanning technology to offer insights and methods that may be used in practical situations to help with proactive monitoring and repair of subway tunnels. The TunnelScanPro framework detected deformations in subway tunnels with 95% accuracy. Safety improved 40% and high-risk deformations decreased significantly, according to the study.
  Keywords: subway tunnelling; deformation monitoring; 3D laser scanning; infrastructure safety; visual analysis; urban transportation; structural integrity.
  
  
• SqueezeNet-deep Kronecker net-based brain tumour classification using MRI image  
  by Srilakshmi Aluri, Sagar Imambi Shaik 
  Abstract: Brain tumour diagnosis is a time-consuming process that heavily depends on the expertise and experience of radiologists. Thus, a SqueezeNet-Deep Kronecker Net (Squeeze-KNet) is devised for brain tumour classification. The proposed model provides accurate diagnosis of brain tumour earlier. Initially, the pre-processing is done using an adaptive bilateral filter and then segmentation process is done using O-SegNet. Thereafter, the feature extraction process is done to extract the significant features like spider local image feature (SLIF), pyramid histogram of oriented gradients (PHOG), local vector pattern (LVP), Weber local binary pattern (WLBP), and local Gabor XOR patterns (LGXP). Finally, the brain tumour classification is done using the proposed Squeeze-KNet, which is the combination of deep Kronecker Net (DKN) and SqueezeNet. The proposed network is evaluated using accuracy, true positive rate (TPR), true negative rate (TNR), negative predictive value (NPV) and positive predictive value (PPV) and obtained 92.6%, 91.5%, 91.2%, 90.6%, and 90.1%.
  Keywords: SqueezeNet; deep Kronecker net; DKN; brain tumour; O-SegNet; adaptive bilateral filter; local vector pattern; LVP.
  
  
• A novel heuristically adaptive dual attention-based long short-term memory for intelligent stock market trend prediction model  
  by Anuja Jana Naik, Madanant Jana Naik 
  Abstract: The deep learning method is designed for the stock market trend prediction through this paper. At first, stock market data are acquired from benchmark sources and are offered to the time series data formation phase. Deep convolutional temporal network (DCTN) is used here. Later, the attained features are provided to the prediction stage, and effective prediction is made by utilising adaptive dual attention-based long short-term memory (ADA-LSTM). Also, their parameters are tuned with the help of hybrid fruit fly spider monkey optimisation (HFF-SMO) by integrating fruit fly algorithm (FFO) and spider monkey algorithm (SMO) to attain an effective stock market trend prediction rate. Thus, the developed model secures effectively high accuracy rate in stock market trend prediction than existing approaches. Hence, the improved model obtained an effectively high accuracy rate in comparison with stock market trend prediction to existing approaches.
  Keywords: stock market trend prediction; adaptive dual-based long-term memory; deep convolutional temporal networks; DCTNs; hybrid fruit fly monkey optimisation.
  
  
• Hybrid optimised deep residual network with trust parameters for intrusion detection in IoT  
  by Asha Rawat, Harsh Namdev Bhor, Jayprabha Terdale, Varsha Bhole, Anuradha Thakare, Vishal Ratansing Patil 
  Abstract: Security issues are still challenging due to the availability of brilliant skills and hacking tools. Thus, detecting the intrusion in the IoT environment is crucial. Hence, this research introduces a novel optimised deep residual network based on the trust and KDD parameters. Here, an efficient mayfly spider monkey optimisation (MSMO) is proposed for tuning the adjustable parameters of the intrusion detector named deep residual network (DRN), which is modelled by hybridising the social behaviour of the mayfly in the mayfly optimisation algorithm (MA) with the foraging behaviour of the spider monkey based on the fission property of the spider monkey optimisation (SMO) to obtain the global best solution. Here, the trust factors and the KDD Cup features are considered for learning the classifier. The proposed model obtained better performance in accuracy of 0.913, precision of 0.919, false alarm rate of 0.084, and recall of 0.958.
  Keywords: intrusion detection; deep residual networks; optimisation; trust factors; KDD Cup features.
  
  

Special Issue on: Knowledge Extraction and Mining to Enhance Intelligent Information Systems

• MS-ConvNeXt: a deep-learning method for tomato leaf diseases identification  
  by Yunchao Li 
  Abstract: Existing deep learning methods for tomato leaf disease identification are challenged by the multi-scale disease regions and complex backgrounds in tomato leaf images. A network for tomato leaf disease is proposed. In the proposed network, a cross-channel-and-spatial attention mechanism is first introduced in the ConvNeXt block (called A-ConvNeXt block) to avoid interference of invalid features from the complex backgrounds. Then, a multiscale feature mechanism is integrated into the backbone constructed by the A-ConvNeXt block to extract features across multiscale diseases. The fine multiscale and silence features are extracted to address the limitations on tomato leaf diseases. Experimental results on laboratory and natural datasets show that the identification accuracy reached 95.67%, which outperformed many other existing networks in comparison experiments. The proposed network may effectively improve tomato leaf disease identification and provide decision-making information for practical applications in modern agriculture.
  Keywords: tomato leaf disease identification; attention mechanism; multiscale feature mechanism; deep learning.
  
  
• A novel multi-sensor fusion approach for enhanced navigation in autonomous driving  
  by Qinghai Liao, Feiyang Cheng, Ji Yu, Zhengguang Ao, Zhiquan Deng, Liang Huang, Huiyun Li 
  Abstract: The limitations of single-sensor SLAM technologies in addressing the intricate requirements of modern intelligent vehicles have prompted a shift towards multi-sensor fusion SLAM as a prominent area of research. In response, this paper proposes a tightly-coupled SLAM system integrating LiDAR, cameras, and IMUs to boost the location accuracy and mapping capabilities. The system processes multi-sensor data upfront to enable effective backend optimisation. Specifically, it integrates LiDAR odometry directly within the vision-inertial framework as inter-frame constraints to streamline computational complexity. Moreover, to counter the progressive error accumulation typical of odometry-based methods, loop closure detection is incorporated, enhancing the quality of localisation and mapping. The effectiveness is substantiated through experiments on public datasets, confirming its proficiency in accurate positioning and navigation. The experimental results demonstrate that the proposed multi-sensor fusion SLAM system maintains high accuracy and reliability across different speeds and environmental conditions, with improvements in trajectory estimation due to loop closure.
  Keywords: autonomous driving; SLAM; multi-sensor fusion; pose estimation; LiDAR odometry.
  
  

Special Issue on: Multi-modal Information Learning and Analytics on Data Integration

• Design and optimisation of electrical information collection and transmission system based on multimodal information analysis  
  by Jinyin Peng, Xiangjin Zhu 
  Abstract: In order to improve the comprehensiveness of power information collection and the flexibility of transmission systems, this article combines multimodal information analysis to conduct in-depth research on system design and optimisation from the perspective of software and hardware structure, and tests it from four aspects: data collection quality, data transmission efficiency, accuracy, and system security. The results show that in terms of data transmission accuracy, the average power data transmission accuracy test result of the system in this article reaches about 92.93%; the average test result of AC simulation transmission accuracy reaches about 93.11%; the average accuracy test result of working condition data transmission reaches about 91.68%. From the experimental results, it can be seen that under traditional finite network technology, the transmission accuracy test results of the electrical energy data, AC analogue quantity, and operating condition data of the five test nodes are about 83.91%.
  Keywords: electrical information; multimodal information analysis; data collection and transmission; system design and optimisation; power monitoring.
  
  
• Optimising path planning and obstacle avoidance algorithm for electrical robots using multimodal information learning techniques  
  by Yang Qiu, Bo Zhou, Lingxiao Chen 
  Abstract: In response to the problems of poor adaptability to complex environments, low success rate of obstacle avoidance, and low accuracy of path planning in traditional path planning and obstacle avoidance algorithms, this paper uses multimodal information learning technology to optimise the path planning and obstacle avoidance algorithms of electric robots. Compared with traditional obstacle avoidance algorithms, optimising obstacle avoidance algorithms using deep learning techniques in multimodal information learning and constructing obstacle avoidance algorithms based on vision and dynamic programming can effectively improve the success rate of obstacle avoidance for electric robots. The average pathfinding time of the six groups studied in this article is 58.59 seconds, which is 4.94 seconds and 3.21 seconds lower than the average values of the ant algorithm and A * algorithm, respectively; in a dynamic obstacle environment, the obstacle avoidance success rate of the algorithm studied in this paper is 96.67%.
  Keywords: obstacle avoidance algorithm; path planning; multimodal information learning technology; electrified robot; Q-learning algorithm; ant colony optimisation; ACO.
  
  
• Application of differential privacy technology in multi-modal data sharing  
  by Zhihai Lu, Bin Wang, Nuanqing Ouyang 
  Abstract: The advent of data-driven technologies and artificial intelligence (AI) has led to an increasing demand for the sharing and analysing sensitive information. However, the paramount concern of preserving individual privacy poses a significant challenge. Hence, an algorithm named differential privacy in data sharing for AI (PrivShareAI) has been utilised. The objective is to enable secure and privacy-preserving data sharing in AI by implementing differential privacy measures and maintaining a balance between utility and privacy. The data-sharing paradigm uses sensitivity limits, noise-enhanced queries, and a universal, secure architecture enabled by a trusted server to encourage shared learning while maintaining maximum privacy. The proposed models efficiency is evaluated with baseline comparison studies with the following metrics: privacy guarantee, accuracy on varying parameters, privacy-utility trade-off, and privacy loss and accuracy measure.
  Keywords: artificial intelligence; differential privacy; data sharing; accuracy; Gaussian noise; gradients; privacy guarantee.
  
  
• Multimodal fake data detection and filtering using GANs and contrast learning  
  by Yuanjie Zou 
  Abstract: In recent years, artificial intelligence (AI) has become an integral part of online education, improving ITS, online courses, and learning management systems (LMS). Online education is predicting students knowledge acquisition based on clickstream data. The lack of focus on student interaction with the content and quizzes offered in lecture videos is a major hurdle to online education. Therefore, this paper proposes a multimodal fake data detection and filtering-based generative adversarial network (MFDDF-GAN) to predict student performance in online learning. MFDDF-GAN aims to ensure that all material used in online education is authentic, of high quality, has protected users, is effective in communication. This MFDDF-GAN approach takes advantage of the information already included in the click sequences rather than relying on characteristics. The experimental results show that the MFDDF-GAN technique produces actionable insights into learning analytics related to video-watching learning performance and knowledge acquisition.
  Keywords: online learning; generative adversarial networks; GANs; support vector machine; SVM; student learning performance.
  
  
• Multi-modal and multi-objective joint optimal planning of medium voltage distribution system based on genetic algorithm  
  by Yigang Tao, Jing Tan, Min Li, Juncheng Zhang, Chunli Zhou 
  Abstract: In order to understand the multi-objective joint optimisation planning problem of medium voltage distribution systems, research on multi-objective joint optimisation planning of medium voltage distribution systems based on genetic algorithm is proposed. This article first focuses on the problem of uneven equipment utilisation efficiency in medium and low-voltage distribution networks and studies the coordination and coordination between medium-voltage lines and connected distribution transformers. Secondly, based on the electricity consumption characteristics of the user industry, a method for estimating the maximum load of distribution transformers based on industry demand coefficients is studied. Finally, a certain actual power grid is selected as an example for verification. The experimental results show that the coordination planning method proposed in this article can effectively guide the reasonable configuration of medium voltage lines and connected transformers, and provide a scientific method for designing user access planning schemes.
  Keywords: medium voltage distribution; multi-objective collaboration; genetic algorithm; GA.
  
  
• Communication algorithm of parallel database HPDB system based on computer intelligent network and data integration  
  by Zhenhua Dai, Tingting Wu, Jun Li 
  Abstract: Traditional database systems are replaced with high-performance parallel databases with intricate and time-consuming querying and data processing demands. As the need exists for processing queries in multiple distinct relations, the database systems are ideally suited for parallel execution. This paper proposes an AINF-HPDB system designed for a parallel high-performance database (HPDB) system using adaptive intent-based network framework (AINF) computer intelligence architecture. The suggested approach utilises flexible and innovative capabilities to optimise communication interactions within the parallel HPDB system. The proposed idea aims to maximise throughput in the operation of the parallel database system, minimise latency, and utilise the intelligence and adaptability of network components to improve data transfer efficiency. To maximise performance and efficiency during large dataset handling across various sectors, the proposed idea finds application in the financial services industry for trading at high frequencies, telecommunications for managing networks, scientific studies for simulations, and internet of things for data-intensive applications.
  Keywords: high-performance database system; computer intelligent network; adaptive intent-based networking framework; adaptive routing; load balancing.
  
  
• Evaluation of digital twin resource allocation and multimodal information learning in internet of vehicles  
  by Ke Wang, Zunhai Gao 
  Abstract: The construction of modern intelligent transportation infrastructure has brought many inconveniences to transportation due to its large number of vehicles, high traffic density. This article applies digital twins to intelligent transportation devices in the internet of vehicles, and studies and analyses the composition of intelligent transportation IoT systems and the application of digital twins in intelligent transportation. First, the intelligent transportation equipment of different vehicles was tested, and then the digital twin intelligent transportation equipment was used to configure resources. The results showed that the average satisfaction score of the intelligent transportation equipment improved by digital twin technology was 7.73 points, while the average satisfaction score of traditional intelligent transportation equipment was 8.26 points, an increase of about 6.9%. Research has shown that intelligent transportation devices based on digital twin vehicle networking can allocate resources more reasonably, ensure the safety of road vehicles, and avoid traffic accidents.
  Keywords: car networking; digital twins; intelligent transportation; internet of things; IoT; internet of vehicles; IoV.
  
  
• Formal modelling of software security requirements based on improved clustering algorithm and multi-modal information fusion  
  by Tangsen Huang, Zhenhua Dai 
  Abstract: Manual analysis and verification are common means and methods of software security requirements that work at present, but they have the disadvantages of being long-consuming and low efficiency. In this paper, the k-means algorithm was used to distinguish feature points by k-value, calculate the probability of each point being selected as the cluster centre, and then obtain a new cluster number. Using the K-nearest neighbour (KNN) algorithm and spectral clustering principle, a clustering analysis method based on multi-attribute decision-making was constructed, which can better realise target recognition in a complex environment. The paper designed a contrast experiment based on the improved clustering algorithm. The results showed that the enhanced clustering algorithm can better model the software security requirements, this article include enhancing target recognition accuracy in complex environments using the k-means algorithm with variable k-values for clustering, integrating the KNN algorithm with spectral clustering principles for effective identification in complex environments.
  Keywords: clustering algorithm; software security; formal methods; model checking; multimodal information fusion; formal modelling.