International Journal of Intelligent Systems Technologies and Applications (23 papers in press)

Regular Issues

• Path Selection for Intelligent Robot Mobile Obstacle Avoidance based on Variable Step Size Ant Colony Algorithm  
  by Qilong Li, Wei Zhang 
  Abstract: In order to overcome the problems of poor performance, low obstacle avoidance success rate, and long time consumption in traditional intelligent robot obstacle avoidance path selection methods, a path selection method for intelligent robot mobile obstacle avoidance based on variable step size ant colony algorithm is proposed. Using laser radar to scan the environment of intelligent robots and obtain observation data, building a kinematic model and probability grid map of intelligent robots based on the observation data. Combining variable step size ant colony algorithm with the kinematic model of intelligent robots in the probability grid map to achieve obstacle avoidance path selection during movement. The experimental results show that under the application of the proposed method, the intelligent robot did not collide during the movement process, the movement path was the shortest, the maximum obstacle avoidance success rate of the intelligent robot was 99.12%, and the minimum path selection time was 0.63 s.
  Keywords: variable step size ant colony algorithm; intelligent robot; obstacle avoidance; path selection; probability grid map.
  DOI: 10.1504/IJISTA.2025.10068141
   
  
• Analysis of Interference Cancellation under Limited Frequency Band Resources for B5G Communication System Application Scenarios  
  by Xin Wang, Na E, Ziyu Wang, Meng Wu 
  Abstract: In a bandwidth-constrained setting, precise self-interference cancellation in communication systems poses challenges. This research extends optical self-interference cancellation to propose an adaptive interference mitigation strategy tailored for advanced mobile communication systems beyond the fifth generation. By considering delay gains in reference and self-interference channels, along with multipath temporal aspects, an unconstrained optimisation algorithm is introduced. Additionally, an adaptive mitigation strategy for multipath channels is devised using orthogonal frequency-division multiplexing technology. Simulation experiments demonstrate the optimised algorithm's 29.8238.15% reduced sampling energy consumption compared to alternatives. The self-interference cancellation strategy saves energy, achieving 30.02 dB detection depth. Post-cancellation, the signal spectrums error vector magnitude is only 6.71%, a 71.72% decrease. The results highlight the effectiveness of this adaptive interference mitigation strategy in self-interference and multipath signal elimination in communication systems.
  Keywords: band-limited resources; interference mitigation; unconstrained optimisation algorithm; delay gains; multipath signals.
  DOI: 10.1504/IJISTA.2025.10068592
   
  
• Intelligent Recognition of Foul Action of Track and Field Athletes based on Bouguet Stereoscopic Correction method  
  by Hao Wu 
  Abstract: This article proposes a new intelligent recognition method of foul action of track and field athletes based on Bouguet stereoscopic correction method, aiming to shorten recognition time and improve recognition accuracy. This method first utilizes the principle of binocular vision to collect images of foul action by track and field athletes, and uses the Bouguet algorithm for image correction to eliminate distortion. Then, the background subtraction method is used to extract the characteristics of foul action by track and field athletes. Finally, the extracted features are input into a support vector machine for intelligent recognition. The experimental results show that the recognition time of this method in multiple data tests does not exceed 1.0s, and the average recognition accuracy reaches 94.388%. This method provides an effective solution for quickly and accurately recognizing foul action by track and field athletes.
  Keywords: Bouguet stereoscopic correction method; Track and field athletes; Foul actions; Intelligent recognition.
  DOI: 10.1504/IJISTA.2025.10068745
   
  
• Brain Tumour Segmentation for Overall Survival Prediction  
  by Novsheena Rasool, Javaid Iqbal Bhat 
  Abstract: Gliomas present significant challenges due to their heterogeneous and infiltrative nature, making accurate segmentation essential for effective treatment. Manual segmentation methods are highly labour-intensive and often inadequate. This study introduces a novel pipeline for improving glioma management, beginning with advanced MRI pre-processing. We propose two attention-gated UNet architectures, the dual convolution attention gated UNet and the channel attention gated UNet, for precise tumour segmentation. Radiomic features, including the grey-level co-occurrence matrix and grey-level dependence matrix, are extracted to capture detailed tumour characteristics. Clinical data, such as age and resection status, are integrated alongside radiomic features to enhance survival models. A stacking ensemble model, combining a random forest regressor and multilayer perceptron, predicts survival based on integrated data. Validation on the BraTS 2018 dataset shows that dual convolution attention gated UNet excels in both segmentation accuracy and survival prediction, highlighting the potential of these advanced technologies for glioma management.
  Keywords: Gliomas; Segmentation; Dual Channel Attention gated UNet; MRI; Channel Attention gated UNet; Survival prediction.
  DOI: 10.1504/IJISTA.2025.10069003
   
  
• Detection of Waterlogging in Urban Road Traffic Based on Improved YOLOv5-seg and Ellipse Fitting Algorithm  
  by Jianqiang Liu, Rui Chen, Xiaoyan Zhao, Xingyao Li, Yujie Shang, Peng Geng 
  Abstract: This article proposes an innovative method for acquiring precise waterlogging depth data utilising images from traffic surveillance systems. Initially, the YOLOv5 algorithm identifies the vehicle type and determines its tire specifications accordingly. Subsequently, an enhanced version of the YOLOv5-seg model segments and masks the tire instances, while an ellipse fitting algorithm extracts the geometric parameters of the submerged tires to shape a complete ellipse. With the vehicle tires as benchmarks, a mathematical model for waterlogging depth is formulated, which computes the depth using crucial parameters from the ellipse. The experimental outcomes demonstrate that this algorithm achieves an average localisation accuracy of 96.4%, a mask segmentation accuracy of 95.6%, and maintains a detection error within 5 cm for 90% of the waterlogged depths measured. These findings confirm that the image-based tire detection method for waterlogging measurement is both effective and practical.
  Keywords: waterlogging depth detection; deep learning; ellipse fitting; YOLOv5-seg.
  DOI: 10.1504/IJISTA.2025.10069443
   
  
• Control of Camless Electrohydraulic Valvetrain in Internal Combustion Engine under Extended Kalman Filter  
  by Lei Zhang, Xiaoqin Yang 
  Abstract: In order to solve the problems of low success rate, long response time of control signals, and high failure rate in traditional valvetrain control methods, a control method of camless electrohydraulic valvetrain in internal combustion engine under extended Kalman filter is proposed. Build a mathematical model for the camless electrohydraulic valvetrain of internal combustion engine, combine the constructed mathematical model with extended Kalman filtering to estimate the state of the camless electrohydraulic valvetrain, and determine the relevant control signals. Input the control signals into the neural network PID controller to camless electrohydraulic valvetrain control of internal combustion engine. The experimental results show that the control success rate curve of the proposed method varies from 95.3% to 98.1%, the average response time of the control signal is 39.09ms, and the failure rate varies from 0.8% to 3.1%, demonstrating high precision and efficiency.
  Keywords: Extended Kalman Filter; Internal combustion engine; Camless electrohydraulic valvetrain; Mathematical model; Neural network PID controller.
  DOI: 10.1504/IJISTA.2025.10069474
   
  
• Anti Tampering Method of Private Data in Industrial Internet for Intelligent Manufacturing  
  by Yafan Men, Jie Lu 
  Abstract: To improve the integrity of industrial Internet privacy data and reduce the tamper response time, an intelligent manufacturing oriented industrial Internet privacy data tamper prevention method was proposed. Firstly, in the context of intelligent manufacturing, through association rule mining methods, analyse the characteristics and relationships of industrial Internet privacy data sets, and mine industrial Internet privacy data. Secondly, through sparse fraction method and L1 norm minimisation strategy, key features of industrial Internet privacy data are extracted, and feature selection process is optimised to improve the accuracy and efficiency of data processing. Finally, by deploying monitoring scripts, encryption processing and key generation algorithms, an industrial Internet privacy data tamper prevention system is built to ensure data integrity, improve security, and prevent unauthorised tampering. The experimental results show that compared to existing tamper proof methods, the data integrity of our method is higher and the response time is the shortest.
  Keywords: Intelligent manufacturing; Industrial Internet; Privacy data; Anti tampering.
  DOI: 10.1504/IJISTA.2025.10069477
   
  
• A Dynamic Resource Allocation Method for Edge Computing of Industrial Internet of Things based on Bee Colony Algorithm  
  by Yanrong Wang, Xiaokun Huang, Ying Li 
  Abstract: In order to solve the problems of poor stability and low efficiency of industrial IoT resource allocation system, a dynamic resource allocation method based on bee colony algorithm for industrial IoT edge computing was proposed. First of all, an architecture integrating the advantages of cloud computing and edge computing was built. Secondly, an energy consumption calculation model was constructed and optimization objective functions and constraints were set. Then, the bee colony algorithm is used to optimise the dynamic resource allocation of edge computing. This algorithm simulates the honey harvesting behaviour of bees and efficiently searches for the optimal resource allocation plan in the solution space through the collaborative effect of hiring bees, observing bees, and reconnaissance bees. Experimental results have shown that the proposed method consistently maintains a throughput of over 45Mbit/s, with high efficiency in processing task requests and resource allocation, and better system stability.
  Keywords: Internet of Things; Dynamic resource allocation; Edge computing; Bee Colony Algorithm.
  DOI: 10.1504/IJISTA.2025.10069479
   
  
• Trajectory correction control method for autonomous mobile robots based on embedded laser ranging  
  by Lei Zhang, Baochen Yang, Wenlian Guo 
  Abstract: In order to reduce the movement deviation of robots, a trajectory correction control method for autonomous mobile robots based on embedded laser ranging is proposed. Firstly, a two-dimensional embedded laser rangefinder is used to obtain point cloud data of the robot's movement, and the robot's movement distance is analysed. Next, proceed with developing the robot's kinematic model and formulating the generalised dynamic equations. Ultimately, the adaptive Monte Carlo algorithm is applied to precisely determine the robot's location and compute the variance between the robot's coordinates and the primary point. Leveraging the classic PID algorithm, a control system for correcting distance and angle is established to enable precise trajectory adjustments for the robot. Notably, empirical findings demonstrate a remarkable 98.4% success rate in robot positioning using the proposed method, while discrepancies in the X-axis, Y-axis, and yaw angle closely correspond to the true values.
  Keywords: embedded laser ranging; autonomous mobile robots; trajectory correction control; kinematic model.
  DOI: 10.1504/IJISTA.2025.10067676
   
  
• A specific action pose recognition of hierarchical dance based on pose feature matching  
  by Yu Zhang, Jun Wang 
  Abstract: To enhance the traditional method's limitations of low accuracy and prolonged feature matching times for specific dance action posture matching, we introduce a hierarchical approach for dance-specific action posture recognition. Initially, we utilise Kinect devices to capture real-time data and extract pertinent physical features. Subsequently, the K-means clustering algorithm is employed to extract keyframe features from the sequence, followed by image reconstruction using the active contour lasso method. Next, hierarchical dance movements are identified through two-dimensional manifold analysis, which enables us to derive the distribution function of edge contour features. Finally, the posture feature matching method is applied to align the functional outcomes, leading to recognition of specific action postures. Experimental results demonstrate that this method achieves a pose feature matching accuracy of 99.8% while reducing the matching time to 1.5 seconds. This method improves the performance of recognising specific movements and postures in graded dance.
  Keywords: active contour lasso method; two-dimensional manifold analysis; postural features; k-means clustering algorithm; feature matching.
  DOI: 10.1504/IJISTA.2025.10067674
   
  
• Texture detail enhancement method of 2D animation image based on generative adversarial network  
  by Zhigang Yang 
  Abstract: To address the issues of poor performance, low PSNR, and long processing time in traditional 2D animation image texture detail enhancement methods, a texture detail enhancement method of 2D animation image based on generative adversarial network is proposed. Build a two-dimensional animation image acquisition architecture using image acquisition modules, IIC driver configuration modules, etc., to achieve two-dimensional animation image acquisition. Extract the features of collected 2D animation images from multiple perspectives such as brightness, local contrast, and saturation, and combine them with generative adversarial network to enhance the texture details of 2D animation images. The experimental results show that the two-dimensional animated images processed by the proposed method do not exhibit any artefacts or distortions, ensuring the realism and naturalness of the images. The average PSNR of the two-dimensional animated images is 53.25 dB, and the average enhancement time is 1.35 s. The image texture details are enhanced effectively.
  Keywords: generative adversarial network; 2D animation image; texture detail enhancement; image acquisition architecture; brightness; local contrast; saturation.
  DOI: 10.1504/IJISTA.2025.10069481
   
  
• Study on internet of things anomaly data mining method based on improved differential evolution automatic clustering  
  by Aihua He 
  Abstract: Due to the diversity of IoT devices and the complexity of the environment, it is difficult to effectively and accurately extract data features, resulting in poor clustering performance, poor accuracy in IoT anomaly data mining, and inability to effectively ensure the security of IoT systems. Therefore, a research on IoT anomaly data mining method based on improved differential evolution automatic clustering is proposed. Through dimensionality reduction of the collected data, the overfitting problem of the mining results is avoided and the mining efficiency is improved. The multi-stage feature selection is expanded to obtain the best feature. Based on this, a differential evolution algorithm is introduced to determine and adjust cluster centres by improving variation factors and cross factors through adaptive strategies, and the K-means automatic clustering algorithm is used to complete abnormal data mining in the network. The results show that the NMI value and ARI value of the proposed method can reach more than 0.95, the AUC value is close to 1, and the mining time is 1.8 s, which has a good clustering effect and can accurately realise the mining of abnormal data.
  Keywords: internet of things; IoT; abnormal data mining; multi-stage feature selection; improved differential evolution; automatic clustering.
  DOI: 10.1504/IJISTA.2025.10067675
   
  
• Multi-dimensional evaluation method of Chinese online teaching effect based on learning behaviour data  
  by Bingxin Zhao 
  Abstract: In order to overcome the problems of low recall and precision of learning behaviour data, as well as low evaluation accuracy in traditional multi-dimensional evaluation methods for Chinese online teaching effectiveness, a new multi-dimensional evaluation method of Chinese online teaching effect based on learning behaviour data is proposed. Using K-means algorithm and AdaBoost algorithm to mine learning behaviour data, Chinese online learning state recognition is performed based on the mined learning behaviour data and Kalman filtering. Combining the results of Chinese online learning state recognition with Markov chain, multi-dimensional evaluation of Chinese online teaching effectiveness is achieved. The experimental results show that the average recall rate and precision rate of the learning behaviour data of the proposed method are 96.78% and 97.34%, respectively. The accuracy of multi-dimensional evaluation of the effectiveness of Chinese online teaching varies within the range of 93.8% to 97.3%, indicating high accuracy.
  Keywords: learning behaviour data; Chinese online teaching effectiveness; multi-dimensional evaluation; Kalman filtering; Markov chain.
  DOI: 10.1504/IJISTA.2025.10067670
   
  
• Edge extraction method for multi-frame 2D animated images based on spatiotemporal filtering  
  by Ting Zhang, Xiang-yu Wei, Feng Qu 
  Abstract: In order to reduce the edge localisation error of animated images and improve the accuracy of edge extraction, a multi-frame 2D animated image edge extraction method based on spatiotemporal filtering is proposed. Firstly, the spatiotemporal filtering algorithm is used to divide the regions to be enhanced in multiple 2D animated images into basic layers and detail layers, thereby enhancing the animated images. Secondly, the Canny operator and an extended Sobel gradient template are used for gradient calculation. Finally, by improving the adaptive threshold segmentation method and combining histogram analysis with the strategy of maximising inter-class variance, the edge extraction process of multi-frame 2D animated images is optimised to improve the accuracy and robustness of edge detection. The experimental results show that the signal-to-noise ratio of the method proposed in this paper remains above 35 dB, and the edge extraction accuracy reaches 99%.
  Keywords: spatiotemporal domain filtering; multi-frame 2D animated images; edge extraction; Sobel gradient template.
  DOI: 10.1504/IJISTA.2025.10067669
   
  
• Adaptive identification and warning method for financial fraud behaviour based on deep Q-learning  
  by Qin Wang, Wei Qing Diao, Fei Ren, Yi Heng Zhang, Mary Jane C. Samonte 
  Abstract: In order to improve the accuracy of identifying financial fraud and reduce the false alarm rate, a new adaptive identification and warning method for financial fraud based on deep Q-learning is proposed. Firstly, analyse the extraction of financial behaviour characteristics, including transaction funds, networks, cycles, and supervised features. Secondly, deep Q-learning combines deep neural networks and reinforcement learning to automatically extract key features from financial transaction data, and continuously optimise strategies through interaction with the environment to achieve accurate identification of financial fraud behaviour. Finally, based on the deep Q-learning model, financial fraud behaviour is assessed for risk and classified into warning levels, and corresponding warning response strategies are implemented according to different risk levels. The experimental results show that the adaptive identification accuracy of financial fraud behaviour using our method can reach up to 98.9%, with a maximum false alarm rate of around 1%.
  Keywords: deep Q-learning; financial fraud behaviour; adaptive discrimination; warning methods.
  DOI: 10.1504/IJISTA.2025.10069608
   
  
• Tackling cyberbullying: a multilingual approach to cyberbullying detection in India  
  by Shahwar Nawshad, Umar Farooq, Parvinder Singh, Surinder Singh Khurana, Anam Bansal 
  Abstract: In India's evolving digital world, women are particularly vulnerable to cyberbullying due to differences in education, limited digital literacy, and pervasive cybersecurity risks. This research focuses on creating a system to detect cyberbullying targeting Indian women. Acknowledging the country's linguistic diversity, we adopt a multilingual approach, constructing a dataset incorporating English, Hindi-English swear words, common Indian slang, and offensive lexicons. We apply various machine learning algorithms to classify cyberbullying incidents. Upon evaluating the results, the relevance vector machine (RVM) algorithm emerged as the most effective, achieving 82.61% and 84.82% accuracy scores in detecting cyberbullying over English and Hinglish datasets, respectively. These findings provide crucial insights for crafting strategies to safeguard Indian women in the digital space. However, more sophisticated and hybrid models are planned for the future to address image, video, and audio-based cyberbullying against women.
  Keywords: cyberbullying; multilingual classification; machine learning; relevance vector machine; RVM; SVM.
  DOI: 10.1504/IJISTA.2025.10067745
   
  
• A method for capturing English oral pronunciation errors based on residual networks and gated convolutional networks  
  by Mingxia Jiang, Yao Zhao 
  Abstract: To address the issues of low classification accuracy and poor capture accuracy in detecting English oral pronunciation errors, this study introduces a novel approach leveraging residual networks and gated convolutional networks. Initially, a multimodal English spoken pronunciation corpus is established by converting annotated corpus data. Subsequently, Mel frequency cepstral coefficients are employed to extract pronunciation features, taking into account the human ear's sensitivity to frequency. A recognition network architecture is then devised, which utilises gated convolutional networks to process continuous video frames, thereby extract spatial and temporal features, and incorporates temporal attention for sequence learning. A classification model is subsequently built upon residual networks, and pronunciation error features are trained, with the capture results being computed via a loss function. Experimental outcomes reveal that the highest detection accuracy of our approach stands at 99.8%, underscoring its high efficacy in capturing English oral pronunciation errors.
  Keywords: residual network; gated convolutional network; oral pronunciation; coarse-grained spatial features; timing characteristics.
  DOI: 10.1504/IJISTA.2025.10068082
   
  
• Collaborative lane changing method for multiple autonomous vehicles in the internet of vehicles environment  
  by Zhang Cheng 
  Abstract: The collaborative lane changing method innovatively utilises real-time communication and data sharing mechanisms, greatly enhancing the safety of the lane changing process and enabling vehicles to accurately predict each other's intentions and behaviours. Via real-time data sharing and system cooperation, the relationships of autonomous vehicles across global, local, and relative coordinates are thoroughly examined to develop a 3-DOF dynamic model. This model precisely captures lane-changing dynamics, laying a robust groundwork for optimising collaborative lane-changing strategies. By defining start and end constraints, and integrating motion states with crash analysis, the synchronised lane changes of multiple autonomous vehicles are achieved. The experimental results show that the collision avoidance success rate of this innovative method is as high as 98%, and the lane changing time does not exceed 3 seconds, demonstrating excellent performance.
  Keywords: internet of vehicles environment; autonomous vehicles; AVs; collaborative lane change; dynamic model.
  DOI: 10.1504/IJISTA.2025.10068083
   
  
• An intelligent detection of defects in underground water supply pipelines based on multi feature fusion and improved SVM  
  by Weishan Chen, Peng Gao, Zhigang Zhou 
  Abstract: To enhance the precision and speed of detecting flaws in water distribution conduits, a sophisticated detection approach leveraging multi-feature integration and an enhanced SVM for subterranean water supply pipeline imperfections is introduced. Initially, the KT-965CCTV pipeline inspection robot is employed to capture endoscopic visuals of underground water conduits. Subsequently, SIFT scale-invariant characteristics, GLCM texture descriptors, and Hu's invariant moment geometric attributes are extracted from the pipeline imagery. The K-means algorithm constructs a visual lexicon, and through sequential integration, these features are amalgamated. The SVM framework is then refined using a binary tree structure, facilitating the establishment of an optimal classification boundary and decision-making function. The fused feature outcomes are fed into the refined SVM to enable automated detection of pipeline anomalies. Experimental data indicate that our methodology maintains a defect detection accuracy exceeding 94% while decreasing the duration of the detection process.
  Keywords: multi feature fusion; support vector machine; improve SVM; underground water supply pipelines; intelligent defect detection.
  DOI: 10.1504/IJISTA.2025.10069473
   
  
• Autonomous obstacle crossing control of wall climbing robot based on fuzzy iterative Q-learning  
  by Hong Zhang 
  Abstract: In order to overcome the problems of low robot positioning accuracy, low success rate of autonomous obstacle crossing, and long response time in traditional methods, an autonomous obstacle crossing control method of wall climbing robot based on fuzzy iterative Q-learning is proposed. Build a dynamic model of the wall climbing robot based on the forces acting on its wall movement, and use the current pose in the dynamic model as the system state vector to locate the wall climbing robot using particle filtering. Based on the results of the wall climbing robot and the Q-learning algorithm, a fuzzy iterative Q-learning controller is constructed to achieve autonomous obstacle crossing control of the wall climbing robot. Experimental results show that the maximum absolute error of the proposed method for wall climbing robot positioning is 1.9 mm, the maximum success rate of autonomous obstacle crossing is 98.4%, and the minimum response time is 0.85 s.
  Keywords: fuzzy iterative Q-learning; wall climbing robot; autonomous obstacle crossing control; dynamic model; particle filtering; Q-learning algorithm.
  DOI: 10.1504/IJISTA.2025.10070147
   
  
• Video stream safety helmet recognition method based on improved faster R-CNN  
  by Xian Yang, Longjin Chen, Zongyi Wang, Shuyu Lin, Danqing Luo, Ziwen Cai, Yuxin Lu 
  Abstract: In order to improve the recognition accuracy and speed of safety helmets, a video stream safety helmet recognition method based on improved Faster R-CNN is proposed. By using the weighted average method to greyscale video stream images, the processing flow is simplified, and median filtering is used to remove noise and improve image quality. The improved Faster R-CNN significantly improves the accuracy and robustness of object recognition by combining the PISA module and deformable convolutional layer (DCN). The PISA module optimises sample weight allocation through importance-based sample weights (ISR) and CARL loss functions. DCN enhances the accuracy of feature extraction by learning the offset of receptive field sampling points. The experimental results show that the safety helmet recognition accuracy of our method reaches 97%, and the recognition speed reaches 30FPS. The improved Faster R-CNN can effectively address the challenge of improving target recognition accuracy in complex scenarios.
  Keywords: improved faster R-CNN; video stream; recognition of safety helmets; sample weight.
  DOI: 10.1504/IJISTA.2025.10070146
   
  
• An intelligent recognition method of athletes' human body movements based on PCA-LBP algorithm  
  by Ran Liu 
  Abstract: To achieve accurate recognition of athletes' movements during different sports processes, an intelligent recognition method for athletes' human body movements based on PCA-LBP algorithm is proposed. Firstly, CCD sensors are used to capture athletes' movement images, and the images are subjected to discrete wavelet transform enhancement processing. Then, the bilateral filtering method is used to remove the noise. Subsequently, we employ the local binary patterns (LBP) technique to capture the distinct characteristics of athletes' body motions, followed by dimensionality reduction utilising principal component analysis (PCA). By meticulously scrutinising the sequential attributes of both training and test datasets, we ascertain whether the detected motion patterns align with those from the training cohort, facilitating intelligent identification. The empirical outcomes underscore the method's efficacy in significantly enhancing image clarity, bolstering the precision of athletes' body motion recognition to a level consistently exceeding 96%.
  Keywords: PCA algorithm; LBP algorithm; athletes; human body movements; intelligent recognition.
  DOI: 10.1504/IJISTA.2025.10070150
   
  
• Improving accuracy of recommendation system using neighbourhood-similarity prediction with noise correction  
  by Kausar Attar, Ashish Jadhav 
  Abstract: Recommender systems are crucial in e-commerce, analysing user preferences for personalised recommendations. As technology advances, RSs are expected to become more sophisticated, but they face challenges like human errors and natural noise. All this natural noise (NN) data has an implicit impact on RS because users' personal preferences and behaviours contribute to natural variations or noise in the rating process. Removing the NN information may result in the loss of potentially valuable data and deviations in prediction results, which are major problems that need to be addressed. In this paper, we propose a method named 'RS-NSP', which combines noise correction (NC) and neighbourhood-similarity prediction (NSP) to manage NN in recommendation systems. The NC initially performs a classification of ratings and later suggests a correction methodology. The corrected rating is recommended based on the NSP technique, where neighbourhood intersection is performed with a similarity measure by characterising items and users. Experiment evaluation with the Movielens-100K dataset shows smaller MAE and RMSE. The result shows an increase in the F1 value measure by 6.7%, suggesting a significant improvement in recommendation quality due to effective prediction and correction of noise ratings.
  Keywords: noise correction; neighbourhood-similarity; natural noise; recommender systems.
  DOI: 10.1504/IJISTA.2025.10070153