Advances in Communication, Devices and Networking

Network on Chip (NoC) performance degradation is significantly impacted by network congestion. Routers using adaptive algorithms, which are capable of detecting traffic congestion, can enable NoC to support complex and dynamic applications. Additionally, there is a growing need in NoC for routing algorithms that can be modified to provide minimal paths and congestion monitoring. The proposed high-speed odd-even congestion-aware routing algorithm for NoC architecture is based on the odd-even turn model and regional congestion awareness. In terms of delay and throughput, the suggested technique performs better than the current algorithm.

For the first time, a quantum-inspired comprehensive study on the low power analog performance of a unique AU-TFET in two distinct technological nodes under the two supply voltages, VDD and VDD/2, has been made in this work. Light is also thrown on the significance of threshold voltage (in the very context of this study) and the method of its extraction. The overall exploration stems from the idea of verifying the orthodox understanding regarding the fact that the down-scaling of a device degrades its analog performance. Interestingly, in this study, it has been found that, unlike the conventional MOSFETs, the aforementioned perception or understanding cannot apply to the projected AU-TFET device in a very simple and straightforward manner.

A new metal oxide glassy ceramic system 0.3Li2O–(0.7) (0.5MoO3–0.4V2O5–0.1ZnO) and its crystalline counterpart have been prepared by melt quenching and a slow cooling route, respectively. Over a broad temperature and frequency range, the temperature and frequency-dependent conductivity of both nanocomposite samples have been studied. Conductivity, depending on temperature, is well established using Jonscher’s universal power law, and the frequency dependency is predicted by Almond–West formalism. The DC conductivity (σdc), hopping frequency (ωH), and frequency exponent (n) have been figured out. The high-frequency dispersion in the spectra of AC conductivity is indicated by the frequency exponent (n) value, which represents the dimensional motions of polarons or charge carriers. The movement of polaron with the energy level in the optical band gap is primarily responsible for the predictable values of activation energy corresponding to AC and DC conductivity (Eac and Edc) and the free energy of polaron migration (EH). The complex impedance (cole–cole) plots defend the nature of conduction observed from AC conductivity spectra. The dielectric properties have been investigated in a wide range of temperatures and frequencies by plotting the frequency-dependent dielectric constant εʹ (real) and dielectric loss εʺ (imaginary) which ensures an understanding of the origin and nature of losses occurring in these materials.

Under the framework of density functional theory (DFT), computational analysis of Fe-doped ZnO as a sensing material has been presented in this paper. The electronic and adsorption properties, including charge transfer, density of states, adsorption energy, band structure, and recovery time, have been studied. The electrical activity of Fe-doped ZnO for the adsorption of H2S and NO2 gases has been studied. Fe–ZnO (replaced O) and Fe–ZnO (replaced Zn) were determined to have a binding energy of − 2.2 eV and − 2.6 eV, respectively, indicating that the Zn-atom replaced by the Fe atom exhibited a more stable structure. The adsorption energy for NO2 gas (− 2.72 eV) was better than that of H2S gas (− 0.55 eV). This means that the material is more effective in adsorbing NO2 gas than H2S gas. The recovery time, or the time the material took to return to its initial state after adsorption was also investigated with different temperatures for H2S–Fe–ZnO and NO2–Fe–ZnO. The band gap of 2.47 eV, 1.28 eV, 0.83 eV and 1.94 eV was calculated for ZnO, Fe–ZnO, H2S–Fe–ZnO, and NO2–Fe–ZnO, respectively. The results suggested that the material has potential for use in H2S gas sensing applications.

With the continuous development of the electronics industry in recent years, electronic products, as a ubiquitous consumer product, have been updated and replaced more frequently. They are becoming lighter, thinner, and more convenient electronic products, and the requirements for production capacity and product quality are gradually increasing. As the core of electronic products, printed circuit boards (PCBs) possess high integration and density, while also requiring high efficiency and quality in production. Any small error in the process of PCB welding and assembly can greatly affect the normal use of electronic products. In this environment, traditional manual inspection cannot meet the increasing demand due to its slow detection speed and easy omission. Therefore, this article will study the defect detection process in PCB assembly and welding based on computer-aided technology. Enhance the PCB public dataset through BigGAN network, use computer data models to detect defect locations and identify types in the bare board dataset, and verify the effectiveness and accuracy of this detection through experiments.

With the growing advancement of artificial intelligence, its scope can also be expanded towards wildlife protection without harming the ecological balance that is caused with the use of invasive techniques. These invasive techniques include fitting of trackers, injecting biochemical transmitters, etc., which not only hinders the physical ecology of the animals but also requires manual searching and fitting the devices. Hence, non-invasive techniques have proved to be very useful in this scenario. The proposed work aims to use non-invasive techniques to identify endangered animals in the wild using YOLOv2 object detection algorithm. Since animal detection in the wild is a challenging task, therefore the captured videos for tracking are converted into a series of image sequences where the detection algorithm is applied. A surveillance system on a smaller scale is proposed for three specific endangered animals based on the detection model that detects and monitors the animals on a real-time basis. The work is carried out for three specific endangered animals, i.e. tigers, elephants, and one-horned rhinos. Since elephants and rhinos can appear similar in terms of skin colour depending on the environmental changes, blob analysis is performed as a support study to resolve this problem. However, it has been noticed that increasing the data also removes this error while using YOLOv2. The model is robust to distortions and unwanted point of references.

In this paper, two second-order generating functions have been suggested to model a class of fractional-order system (FOS) in z domain using direct discretization method. The generating functions named as Goswami et al. operator (Method-I) and Mekhnache et al. operator (Method-II) are employed to obtain the approximated models of fractional-order differentiator (FOD) of one-fourth order and one-fifth order via continued fraction expansion (CFE). Further, these two methods have been used to model another two FOSs taken from the literature. Frequency responses of the approximated models show that Method-I has produced less root mean square error (RMSE) than Method-II and hence appears to be more effective to approximate the FOS.

Many debilitating, fatal disorders, like Parkinson's disease, which are becoming more common, exhibit neurodegeneration. Development of unique, more potent medicinal approaches is crucial in order to fight these deadly diseases. Commercial gait detection systems based on force plates and footprints have been effectively used in the clinical diagnosis of such diseases. The classification of model for old versus young subjects with and without neurodegenerative diseases was constructed using MATLAB software. The analysis with respect to stride interval is mentioned in this paper. The 15 subjects were taken for this experiment; among these, five healthy young as well as old individuals were considered and five older adults with Parkinson’s disease were taken. Therefore, the two classes were formed using SVM kernel modeling for the diseased and healthy subjects. The classification for stride length interval (0.95–1.5 s) for SVM modeling was providing 96.7% validation accuracy.

Recent developments in deep-sea exploration, environmental surveillance, and ocean research, accurate segmentation of underwater images is essential. This study pursues this goal by exploring underwater image segmentation from a deep learning perspective. It specifically looks at how well the Attention Residual UNet architecture, a more sophisticated version of the U-Net, works with the focal loss technique to achieve accuracy in this crucial task. Using attention mechanisms and residual connections, the Attention Residual UNet architecture, based on the U-Net framework, can capture fine details while maintaining contextual coherence. Model loss and accuracy measures are considered as this study carefully assesses the architecture’s performance. Notably, the model exhibits outstanding accuracy, obtaining 97.23% and 92.76% accuracy on the training and validation sets, respectively. The Jaccard coefficient further demonstrates the model’s efficiency, which measures the intersection between predicted and actual segments and has coefficients for the model and validation sets of 74.31% and 66.51% , respectively. The Mean Intersection over Union (MIoU) statistic, which boasts a value 91.24%, validates the model’s superiority.

Advancement in digital pathology has enabled deep learning-based computer vision techniques for automated diagnosis and prognosis of diseases. The essentiality of early detection and prognosis of any cancer category lies in the fact that it can speed up the subsequent medical treatment procedures of patients. About 10% of all cancer cases worldwide are related to colorectal cancer (CRC), and it is also the third most common category of cancer (Egeblad et al. in Dev Cell 18:884–901, 2010). So, it is clinically important to classify and make an objective evaluation of colorectal cancer histological images. The classification performance of current methodologies primarily relies on the use of various combinations of texture-based features and classifiers or transfer learning to classify different organisational kinds. As a result of the diversity of tissue types and characteristics present in histological images, classification is still a challenging task. In this study, we have put forth a novel and effective stacking (Wolpert in Neural Netw 5:241–59, 1992) ensemble (Zhang and Yunqian (eds) Ensemble machine learning: methods and applications. Springer Science; Business Media, 2012) technique for classification on the histopathological image analysis benchmark dataset Kather-5 K (Kather et al. in Sci Rep 6:27988, 2016). The ensemble consists of two cutting-edge deep learning architectures, ResNet18 (He et al. in Proceedings of the IEEE conference on computer vision and pattern recognition 2016, pp 770–778) and EfficientNetB0 (Tan and Le Efficientnet: rethinking model scaling for convolutional neural networks. International conference on machine learning 2019 May 24. PMLR, pp 6105–6114), acting as weak learners, and an ANN acting as the meta-learner. The proposed approach obtained a remarkable accuracy of 97.20% in CRC classification on the said dataset.

Diabetes is a chronic metabolic disorder that affects millions of people worldwide. Early detection and effective management of diabetes are crucial to prevent severe complications and improve the quality of life for affected individuals. Machine Learning (ML) techniques have shown great promise in aiding the early detection of various diseases on patient data, including diabetes. ML algorithms can analyze vast datasets, identify patterns, and make accurate predictions, helping medical professionals to diagnose diabetes at its early stages. In our work, we employed several ML models for diabetes classification using different datasets. These models include K-Nearest Neighbor (KNN), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient-Boosted Model (GBM), eXtreme Gradient Boosting (XG-Boost), Adaptive Boosting (Ada-Boost), Support Vector Machine (SVM), and Gaussian Naive Bayes (GNB). We performed a comparative analysis of their performance on three distinct datasets using evaluation metrics like accuracy, precision, F1-score, sensitivity, specificity and Cohen’s Kappa Value. Our findings revealed that the RF algorithm is optimal for symptoms-based and primary lab report-based diabetes detection, while XG-Boost excels in classifying different types of diabetes from a multi-class dataset. Moreover, we investigated diverse symptoms and their impact on diabetes outcomes, offering insights into preventive measures and early stage monitoring for this disease classification.

Social media attracts all kinds of activities, including product marketing, celebrity marketing, and also it serves as platform for promoting political agenda. As it is gaining popularity from all various source, it has also attracted spammers and automated accounts that are responsible for spreading the misinformation and influencing the audience. In this context, there is a need to properly classify the social media account as bot account or human account. For classification and detection of social bots, different machine learning, deep learning techniques are implemented. In this paper, we have focused on ensemble technique for classification of social bot. Considering heterogeneous base classifier, such as decision tree, logistic regression and k-neighbor classifier, an ensemble model has been built, that combines the prediction of base classifier, and gives the final prediction. The ensemble approach that has been implemented are, majority voting, random forest and bagging with decision tree.

Weather prediction is a vital part of our life. The collective information about the change in temporal dynamics of weather is very significant. This present work summarizes to analyze the current research works on weather forecasting and compare different machine learning techniques that has been used by several researchers for weather prediction. Various input parameters to these machine learning models are also tested to determine the usefulness of each of them. The fundamental aim of this present study is to develop a real time integrated system for weather monitoring system which enables prediction of temperature, humidity, and pressure information. Multivariable linear regression-based prediction model is used here to obtain the appropriate prediction model. Dataset is considered here of a city of tropical country India for the prediction model. It is an existing dataset which has been downloaded from online data repository. This work elaborates the proposed prediction model and the corresponding analytics supported by comparative study with the variation of different parameter. Temperature and humidity is predicted here among different weather parameters. Along with this, a hardware prototype model for real time for weather data acquisition using Raspberry Pi 3B model is reported in this present work. Raspberry Pi will fetch the data from the sensor, and all those data will be stored into a database and display of this data using a GUI is also shown for monitoring purpose.

Using the pre-trained models, this paper discusses the classification of four Khasi dialects—Sohra, Nongkrem, Mairang, and Maram dialects. Mel-spectrogram images were extracted from speech audio of the above four dialects with time masking augmentation. With pre-trained AlexNet and ResNet18 models, we obtained remarkable outcomes in our dialect spectrogram identification study. In our experiment, we got a decent validation accuracy of 93.58%, 93.25%, and 93.20% by AlexNet with 8, 15, and 25 epochs, respectively. Again, with the same epochs, ResNet18 achieved accuracy rates of 88.23%, 93.55%, and 93.57%.

This study focuses on efficient elevation value localization and recognition in topographic sheets (TS) through morphological operations and YOLO-based deep learning. The aim is to enhance the digitization process, crucial for creating Digital Elevation Models (DEMs) widely used in various applications.

Internet facility has led to emergence of patient support groups. These have gained prominence as these fulfils important benefits to patients. One such benefit is patient empowerment. These online groups provide opportunity to patients to interact with similar ailments and predicaments and who can understand the pain and discomfort felt by the patient. This provides validation for the patient and patients’ experiences. How does this homophily impacts patient empowerment? This question has been explored in this study. The methodology is based on an online survey of patients visiting such online platforms. In all 701 patients provided the data. Independent variable (homophily) and dependent variable (patient empowerment) have been measured using a 7-point Likert scale. Findings provide that both are weakly correlated, but this correlation is significant. Regression analysis led to a regression model that is fit statistically. This provides basis to encourage patients to visit online support groups.

Biometric authentication scheme has been widely adopted for authentication purpose due to its uniqueness, universality, and distinctiveness. However, research has shown that these schemes are not necessarily more secured, especially with issues of coercion proliferating the cyberspace. Also, available techniques on repudiation of biometric features for authentication have not adequately explored this exciting topic. Integrating emotional state into the authentication scheme helps to mitigate coercion. This paper presents a framework for emotion as a way of biometric authentication scheme. An emotion classification model was developed by training an emotion classifier brainwave signal from eight subjects under normal state and duress using KNN machine learning algorithm. The authentication scheme grants access to users who pass both verification phases. The emotion classification model achieved an accuracy of 93.6% on the full feature set, while the reduced-feature set, as a result of feature selection, produced an improved accuracy of 94.15%. Hence, the emotion classification model can be integrated into existing biometric authentication system for improved security of critical user information.

Speech recognition is the process of recognizing and understanding human voices, and converting them into textual information. Speech recognition is a complex and important technology, and it is an important research topic. On this basis, a speech recognition method based on embedded technology was proposed. Based on the above research, this article designed and implemented an embedded Japanese translation system. Language recognition technology is a machine that recognizes, understands, and converts language information into textual information. Nowadays, computer dictionaries and sound recognition technology are both advancing. Due to the advancement of chip technology, embedded systems have more functions and speech recognition technology can be embedded into embedded systems. Therefore, embedded speech recognition has become a new development direction. This article explored a Japanese speech recognition and real-time translation system based on convolutional neural networks (CNN). Firstly, it explored how to build a speech recognition system and then constructed a translation system. After that, it constructed an end-to-end speech recognition model. Finally, the superiority of the system in this article was verified through experiments (the accuracy, recall, and F1 mean of the speech recognition and real-time translation system based on the algorithm in this article were higher than or equal to 0.78).

Based on the research on the physical fitness and health of college students, using methods such as system clustering, one-way ANOVA, and discriminant analysis, models for BMI grading, cluster analysis, ANOVA, and comprehensive evaluation of physical fitness and health were established. Software programs such as MATLAB, Eviews, and SPSS were used to analyze the impact of weight and student origin on physical fitness and health issues, verify the accuracy of the test data, and eliminate some erroneous data; finally, an evaluation standard for the physical health of college students was obtained, and this standard was used to evaluate the institutional health status of some of the college students. The experimental results indicate that, in the sample with a population of 29, there are 4 individuals with excellent physical fitness, accounting for 13.79%, 12 individuals with good physical fitness, accounting for 41.38%, 7 individuals with qualified physical fitness, accounting for 24.14%, and 6 individuals with failed physical fitness, accounting for 20.69%. The results obtained during our overall analysis are roughly similar and proved the feasibility of multivariate statistical analysis of national traditional sports physical fitness and health test data.

The article proposes a method for predicting interprovincial environmental efficiency based on the DEA model. Firstly, the characteristics of various DEA models were compared and analyzed, and then the DEA model was used to predict interprovincial environmental efficiency in China through examples. The results indicate that with the continuous improvement of provincial industrial environmental efficiency, the impact of provincial environmental protection investment on environmental efficiency shows significant fluctuations. The fluctuation range is controlled between [0.28, 0.26]. The results also indicate that the more severe environmental pollution, the lower environmental efficiency.

The outdoor games of young children are mainly in autonomous way, which can intuitively reflect the inner thoughts of young children and promote the all-round development of young children, in which the teachers have an important guiding role in the games of young children. Therefore, the study proposes the analysis of autonomous outdoor play guidance strategies based on K-means clustering algorithm, through questionnaire survey and interview observation method to investigate the autonomous outdoor play guidance strategies of 615 early childhood teachers with different characteristics and classify the teachers’ guidance strategies into three types by combining the K-means clustering algorithm. Statistical results found that there were four dimensions of significant differences among the three types of teacher instructional strategies, with the autonomous exploratory type having the highest overall scores, indicating that autonomous outdoor play has a positive effect on the development of young children's physical health and behavioral interactions. The study proposes appropriate teaching guidance for the results analyzed in this survey, giving full play to the role of role models and sharing relevant experiences to learn from others’ excellent experiences and supplement their own teaching shortcomings.

In order to generate high-quality images for expanding data sets and image classification, this paper proposes a method of art image generation system based on artificial intelligence. Combined with the advantages of the current image generation domain model, a two-stage image generation method with intermediate inputs is introduced. The main process of this method is use a basic GAN model to fit the features of the output image through a feature capture network (classification network). The generator with generating classification feature is fused into a basic GAN model to form a new GAN model for generating pictures. The experimental results show that the FID score of the basic model is 8.14, and the FID score of the model in this paper is 5.27. According to the measurement standard of FID, the lower the FID score is, the better the picture quality is. Verified that the two-stage image generation model outperforms the basic model in generating the MNIST dataset, indicating the effectiveness of the model. Conclusion: the model using this method has stronger pattern generation ability than the basic model.

The healthcare industry is increasingly recognizing the need for efficient management of electronic patient records (EPRs). This paper explores the potential of using Hyperledger Fabric (HLF), an open-source blockchain framework, for the adoption of EPRs in the healthcare industry. We investigate the concepts of blockchain, distributed networks, and smart contracts, and analyze how HLF can provide a secure and decentralized platform for storing and managing sensitive medical data. By leveraging HLF’s features such as permissioned networks, chaincode, and private channels, healthcare providers can ensure data integrity, confidentiality, and availability while granting and revoking access to authorized personnel. This paper contributes to the understanding of how blockchain technology, specifically HLF, can revolutionize the healthcare industry and improve the quality of care provided to patients.

Stock market predictions are been made with the advent of technological wonders such as worldwide digitalization, redesigning the traditional methodology of trade. Stock trading has evolved into a major area of investing for many financial investors due to the continuous rise in market capitalization. Many analysts and academics have created methods and systems that forecast stock price changes and aid investors in making wise choices. Researchers can forecast the market using unconventional textual data from social platforms thanks to advanced trading models. Both researchers looking for insights concealed in stock market data, and others looking to make money trading stocks have shown a keen interest in forecasting market movements. It is highly challenging to develop a system that can accurately anticipate the future direction of the stock market due to the tremendously nonlinear character of the stock market data but we have developed a website named TradeWise which is an innovative and autonomous website designed to revolutionize financial markets. Utilizing advanced machine learning and artificial intelligence algorithms, it analyses vast datasets, market trends, and real-time indicators to make informed trading decisions.

‘Fintech’ is a term derived from two words ‘Finance’ and ‘Technology’ and refers to technological innovations dealing in financial services. Fintech is gradually emerging as an important interface for rapid financial transactions. Mobile Payment (M-Payment) wallets are one of the most prominently used Fintech services in India. Several theories propose different variables that impact the adoption and usage of mobile payment wallets. This study in tends to determine a relationship between the values of expressed beliefs on Effort Expectancy (EE) with the values of Performance Expectancy (PE), Self-Efficacy (SE), and Perceived Trust (PT). The EE and PE are the variables of technology adoption given in the Unified Theory of Acceptance and Use of Technology (UTAUT). The study is based on the responses received from Bank Customers in the state of Sikkim, located in the northeast part of India, and is empirically tested and validated. It was observed that the values of EE are significantly higher than the values of the other 3 constructs, i.e., PE, SE, and PT. It can be concluded from this finding that variable EE has more significance in determining the adoption intention of Fintech amongst Bank Customers compared to the other 3 variables.

Intrusion detection in modern networks, encompassing the Internet of Things (IoT), software-defined networking (SDN), and cloud environments, represents a pressing research challenge for network security researchers and practitioners. Our research paper focused on utilizing the UNSW-NB15 intrusion dataset and applied a diverse set of machine learning(ML) models to evaluate their performance in this context. However, the dataset presented unique challenges, being highly imbalanced and featuring nine distinct types of attacks. Consequently, many conventional ML models struggled to accurately identify these attack types with high precision. To address this challenge, we have introduced a novel probabilistic-based method to select class-specific instances and conducted feature analysis to pinpoint the most informative attributes for training ML models. The objective was to equip these models with the capability to provide high-precision detection. The outcome of this endeavour was highly promising: our proposed instance selection method consistently delivered accuracy rates exceeding 99% and 98% across a range of tested ML models, supporting both binary and multi-class classification tasks, respectively. These findings underscore the potential of our approach in enhancing the accuracy and effectiveness of intrusion detection in modern network environments, offering a valuable contribution to the field of network security research.

A key moment in health care is marked by the Fourth Industrial Revolution, commonly referred to as Industry 4.0. This transformation, driven by the convergence of digital technologies with automation and data driving processes, has led to a paradigm shift in how health care is provided. The integration of the emerging technologies in Industry 4.0, such as Internet of Things, Artificial Intelligence, Big Data Analytics and Advanced Robots, are revolutionizing patient care, improving resource allocation and shaping research's landscape. To learn more about the ever-evolving relationship between Industry 4.0 and health care, this research paper begins with a bibliographic analysis. In this interdisciplinary convergence, our bibliometric analyses serve as a lens through which we can see the key trends, research areas and influential players. The review of literature highlights the profound impact of Industry 4.0 on health care, revealing that Internet of Things technologies for real-time patient tracking, proliferation of artificial intelligence in medical diagnosis and transforming power of big data Analytics are changing health care decision making. Methodologically, we leverage bibliometrics as a quantitative analytical tool, drawing on citation counts, bibliographic coupling, and keyword co-occurrence analysis. The data for this analysis, which covered the period 2015–2023, was carefully collected from Scopus database. The analysis of the information reveals that, particularly from 2018 onwards, there has been a significant increase in publications concerning Industry 4.0 and health care. In this research landscape India has emerged as a strong contributor, with countries such as the United States and Italy making significant progress. Publication trends and bibliographic coupling among countries and sources shed light on collaborative networks and research focus. The emergence of machine learning, artificial intelligence and data analysis as important themes is illustrated by a co-occurrence analysis of keywords that elucidates evolving research interests. In the complicated terrain of health care converging with Industry 4.0, this research paper serves as a compass. The report highlights this convergence's transformative potential, highlighting the pivotal role that bibliometrics analysis must play in determining future research areas in adopting Industry 4.0 in the health-care sector.

This report explores the integration of high-end neural network models into embedded systems, its advantages over the high-end computers and Cloud-based inferencing to overcome the drawbacks such as cost, power consumption, latency, security concerns, and reliability. The STM32 microcontrollers are widely used in various embedded applications and are known for their high performance, low power consumption, and extensive connectivity options. A CNN architecture, ConvNet3, is proposed and fine-tuned for object classification and navigating obstacles in real-time scenarios, leveraging the limited resources of the STM32F401RE board. The performance of these models is evaluated, considering factors like prediction speed, accuracy, and model size. The findings highlight the trade-offs between accuracy and size reduction achieved through compression techniques. Overall, this research contributes to advancing CNN-based solutions for embedded systems and offers insights for model selection and data preparation, facilitating improved navigational capabilities and object recognition in resource-constrained environments.

Skin cancer is characterized by the uncontrolled proliferation of abnormal cells in the outermost skin layer, the epidermis, due to unrepaired DNA damage leading to mutations. These mutations cause rapid multiplication of skin cells, forming malignant tumors. The primary types of skin cancer include basal cell carcinoma (BCC), squamous cell carcinoma (SCC), melanoma, and Merkel cell carcinoma (MCC). Melanoma of the skin ranks as the 17th most common cancer worldwide, with more than 150,000 new cases reported in 2020. Early detection and treatment of melanoma can significantly impact patient outcomes. The present work aims to detect melanoma skin cancer in its early stages using image processing through Computer Vision and deep learning methodologies. The culmination of this effort is an Android application designed to facilitate self-diagnosis for users, offering timely alerts on when to consult a medical professional. Hospitals can also utilize the application to prioritize patient care based on their risk percentages, benefiting both patients and healthcare providers. The study delves into relevant research papers published in esteemed journals related to skin cancer diagnosis. Deep learning methods are proposed to assist dermatologists in achieving early and accurate diagnoses. While specialists can provide accurate diagnoses, the development of automated systems becomes crucial to efficiently diagnose diseases, saving lives and reducing healthcare and financial burdens. Machine learning (ML) emerges as a valuable tool in this context. The article focuses on the fundamentals of ML and its potential in aiding skin cancer diagnosis. The objective is to conduct a comparative study between the DenseNet-121, ResNet-50, and CNN-RF models. The study reveals that DenseNet-121 outperformed with a testing accuracy of 83%, surpassing ResNet- 50, which achieved 81% testing accuracy. This comparative analysis contributes to the ongoing research and development in the field of skin cancer diagnosis.

Liver cirrhosis is a prevalent and potentially life-threatening condition characterized by the irreversible scarring of the liver tissue. In this study, we propose a comparative approach employing three deep learning models, Convolutional Neural Networks (CNN), Multilayer Perceptron (MLP) Classifier, and Logistic Regression (LR) to enhance the diagnostic accuracy of liver cirrhosis. The primary objective of this paper was to develop and evaluate an automated prediction system. This system utilizes a comprehensive database of cirrhosis data, with a particular focus on improving the detection of liver cirrhosis. The goal of this study is to examine how well three distinct supervised deep learning models performed when it comes to liver cirrhosis detection utilizing actual inter-patient records. Four key measures were used in the study to assess the models’ performances: F1-score, accuracy, precision, and recall. Finally, comparative analysis has been made to showcase the performances indices of this three deep learning models.

This article mainly designed and implemented an aviation maintenance management system based on Java technology. In the system, it was introduced how the website completes aviation aircraft maintenance information management through code writing. This article first elaborated on the background and significance of the topic, research status, and development trends, and then discussed and analyzed the introduction of airlines, including company overview, route plans, and the MySQL database. Finally, based on the theoretical knowledge of software engineering, the functional module structure was determined, and a detailed description was provided to explain the design process and implementation methods. At the same time, the system functions were tested. The test results showed that the response time of the system was between 2 and 4 s, and the concurrency was between 74 and 79%. The system security was between 85 and 98%, the system stability was between 80 and 89%, and the test case coverage rate was 80% or above.

Objective: To address the problems of dispersed educational information resources and low resource sharing in current education platforms, blockchain technology is applied to education cloud platforms to improve the concentration and sharing of educational resources. Methods: This paper built a smart community education cloud platform based on blockchain technology, and specifically designed the service content of the platform. In order to improve the load capacity of the platform, this paper also tested the performance of the platform in combination with the load balancing algorithm. Result: The experimental results show that under the algorithm proposed in this paper, when 600 users access the platform, the CPU (Central Processing Unit) utilization rate reaches 45.86%, and the memory utilization rate reaches 37.97%. Conclusion: From the above data, it can be seen that the algorithm proposed in this paper can effectively reduce the CPU and memory usage of the platform and improve its operational efficiency when facing a large number of user visits.

With the development of society and the progress of science and technology, design has been everywhere with human life. The focus of design is to help users understand the product and coordinate the relationship among designers, products, and users. Product modeling, as the most intuitive form of external expression, is the first product characteristics observed by people. As an important front-end behavior of product modeling design, design evaluation is not only an effective channel for multi-scheme decision-making and selection, but also a key link affecting the final effect of products. How to guide product design positioning and enterprise development strategy to unify through design evaluation has become the premise of enterprise product upgrading and sustainable and stable development. Based on this, this paper chooses to use computer vision algorithm for product modeling design evaluation. Since the computer vision algorithm will not change with the change of evaluation conditions, this paper constructs the evaluation method combining gray theory evaluation method and computer vision algorithm. Finally, through the verification and analysis of the effectiveness of the system, it is found that the system model constructed in this paper basically converges after 500 rounds of training, indicating that the system test effect is good and meets the requirements of intelligent evaluation of product modeling.

In response to the current problems of insufficient objectivity, lack of intelligent management, and lack of feedback and incentive mechanisms in human resource management, the author proposes a work evaluation system based on artificial intelligence algorithms and deep data analysis technology. Using data mining technology, feature indicators with strong correlation with performance are extracted from employee work indicators, and the extracted indicators are classified using artificial intelligence-related algorithms to complete job evaluation. On this basis, utilizing the transparency and reproducibility of classification algorithms, the evaluation process is outputted while outputting the evaluation level. Using the evaluation process as feedback for employees’ work, achieving a motivating effect on their work. Compared with traditional evaluation systems, this evaluation system has the advantages of strong objectivity and simple operation. In the test, the accuracy rate of the evaluation system reached 94%, thus proving the correctness and effectiveness of the system.

In order to push videos that users are interested in and improve user experience, this paper proposes a personalized video new media recommendation method based on deep neural network algorithm. The algorithm combines deep learning and content-based recommendation algorithm. By constructing a deep network model, it extracts the features of video and users’ text information, completes the distributed feature representation of video and users at the semantic space level, deeply excavates the potential relationship between users and video, and makes video recommendation. The experimental results show that the AUC index of this algorithm increases first and then decreases, while the RMSE index decreases first and then increases. When the learning rate is around 0.03, the AUC and RMSE reach the maximum and minimum values at the same time. This shows that when the learning rate is 0.03, the performance of the system is the best at this time. The AUC index and map index of the recommendation algorithm based on the deep semantic model proposed in this paper are higher than the other three algorithms, while the RMSE index is lower than them. Conclusion: the results of the recommendation algorithm proposed in this paper are better than the comparison algorithm, and can complete the video recommendation task better.

In recent years, cross-border e-commerce platforms have experienced rapid development, and the scale of cross-border goods has continued to expand. With the continuous improvement of relevant policies for cross-border e-commerce import and export in various countries, global logistics capabilities and supply chain security have been highly valued, creating basic service capabilities for external circulation. This article analyzes the concept and characteristics of logistics alliances, and provides four different choices for the operation mode of logistics alliances. This article analyzes the comprehensive efficiency, pure technical efficiency, and scale efficiency of 30 logistics alliances through the DEA method, and proposes strategies to improve the efficiency.

With the development of the power industry, smart grid has become the development direction of the future global power grid. With the help of a new generation of information technology, the collection of power consumption data at the power terminal, and the analysis of power consumption data, and then timely prediction or identification of electrical faults, timely warning to the user or automatic cut off the power supply, can effectively reduce the risk of electrical faults caused by accidents, reduce the loss caused by accidents. Therefore, this paper mainly introduces the development and implementation of intelligent electricity safety monitoring system and fault arc identification algorithm. Firstly, the intelligent electricity safety monitoring system is designed. The main functions of the system are as follows: terminal data collection function, LoRa network wireless transmission function, intelligent gateway data exchange function, server human–computer interaction function and intelligent processing algorithm function. Finally, through the system test, it is found that the intelligent electricity monitoring system can identify the fault arc with 96% accuracy, which is suitable for customer-side safety energy monitoring. Through the research and application of intelligent electricity safety monitoring technology, users can grasp the status information and operation of electrical equipment in real time, and establish a perfect operation and maintenance management system based on intelligent electricity monitoring technology.

In order to correctly observe and report the incidence of mental disorders among college students, an intelligent monitoring and warning system for college students’ mental health based on big data technology was proposed. This study investigated the mental health, learning, and behavior patterns of college students and found the differences in their mental health status between different learning styles and lifestyles. Using sleep, staying up late, socializing with classmates, academic performance, absenteeism, and sudden illness as information collection indicators, establish a psychological crisis monitoring and warning system based on big data technology, conduct two-level monitoring and warning, achieve dynamic tracking and accurate monitoring and warning of college students’ mental health status, and improve the level of mental health education in universities.

In order to improve the information security protection capability of the power system network, the author has designed a compliant grid connection scheme for relay protection equipment based on data security. This scheme adopts isolation protection technology based on logical mapping, secure transmission technology based on data mirroring, and data security assessment technology based on risk matrix method. And through experimental verification of the proposed scheme, the practical results show that it can achieve fast response to the safety status of all recorders in the substation and safe transmission of compatible recording data, basically eliminating the influence of the safety conditions of the recorders themselves, and achieving the goal of operating a 99.9% connection rate for the recorders.

As 5G networks continue to evolve and pave the way for future telecommunication technologies, the role of Artificial Intelligence (AI) and Machine Learning (ML) in optimizing traffic management becomes increasingly crucial. This paper explores the integration of AI and ML in telecommunication networks, focusing on their applications, challenges, and potential solutions for traffic optimization in 5G and beyond. The paper looks into specific use cases, such as network congestion management, quality of service (QoS) enhancement, and energy efficiency improvements. Additionally, the paper discusses the implications of AI-driven traffic optimization on network performance, user experience, and the broader telecommunication industry landscape. Through this review, the paper shed light on the transformative potential of AI and ML in shaping the future of telecommunication networks.

Product recommendation systems are a critical element of e-commerce platforms, as they enable customers to identify the items that best meet their needs. Product recommendation systems are important tools used by organizations to increase customer engagement, satisfaction, and loyalty. Existing product recommendation systems have several weaknesses, such as an inability to incorporate features outside of customer preferences. Therefore, this paper proposes an enhanced product recommendation system using a decision tree algorithm. The main objectives of this system are to improve the accuracy and efficiency of the existing product recommendation systems. To accomplish this, a framework includes a data preprocessing phase, a feature selection phase, and a model training phase. The data preprocessing phase is used to clean the data and eliminate any noise. The feature selection phase is used to identify the most informative features, which are then used to train the decision tree model. The model is trained using CART, a supervised learning algorithm, and is evaluated using various metrics such as accuracy, precision, recall, and F1-score. Finally, the model is tested on a test dataset to compare with existing solutions. The results show that the proposed system outperforms existing recommender systems in terms of all the evaluation metrics discussed. Furthermore, the improved system also provides useful insights into the product recommendation process for future studies.

Keyloggers are malicious software programs that record keystrokes of users without their consent or knowledge. They can steal sensitive information like credit card numbers and passwords. They pose a significant threat to users’ privacy and security since they capture keystroke data. Keylogger detection systems play a vital role in safeguarding users from cyberattacks and mitigating the potential harm caused by keyloggers. Keylogger detection systems employing deep learning have been widely used for identifying and mitigating cyber threats associated with keyloggers. However, they struggle to identify new or unfamiliar keylogger samples. This study aims to explore deep learning techniques with feature selection model to detect keylogger. This study employs Recurrent Neural Networks (RNNs) using a collection of known keylogger dataset. Furthermore, a correlation-based feature extraction method was applied to identify the most relevant features for the model, highlighting the importance of specific features, such as URG flag Count, ACK flag count, and Idle Mean, in differentiating between benign and keylog classes. The proposed model demonstrates superior accuracy of 0.8763 and precision of 0.8569 compared to a baseline model using Logistic Regression (LR) 0.7721 and 0.7407, respectively, indicating a better balance between accurate classification and minimizing false negatives and false positives. The findings from this study help to improve keylogger detection methods, making them more powerful and efficient. These results suggest that a keylogger-based detection system employing a deep learning approach can be a valuable tool for addressing the complex and evolving landscape of cyber threats related to keylogging.