Signal Processing, Telecommunication & Embedded Systems: AI and ML Applications

The world’s demand for energy is consciously on the rise. Renewable energy sources have numerous advantages over conventional fossil fuels. Although they provide many benefits, such as sustainability and ecological advantages, the practical integration of these sources into traditional grid systems faces numerous challenges, given that they are relatively new technologies. One of the main challenges is the intermittent nature of energy production from renewable resources. Employing artificial intelligence (AI) in time series forecasting can represent a potential solution to the challenges connected to balancing the production and consumption of energy from renewable resources. This paper aims to employ long short-term memory (LSTM) networks to forecast energy production from renewable sources, specifically solar power plants. To achieve satisfactory performance, metaheuristics were applied for hyperparameter optimization. Additionally, this paper presents an adjusted type of a popular optimization algorithm, aiming to overcome the limitations of the original approach. Two experiments were conducted on publicly available datasets. The optimized models show good results, with an error of less than 1 V in predictions for one hour ahead.

CMOS technology is essential to the development of VLSI systems in today’s world. In deep submicron CMOS technology, power optimization has taken precedence over other considerations. The main issues resulting from the device’s shrinkage are power consumption reductions and overall chip power management. Optimization of power is a crucial for several designs in order to lower package costs and enhance battery life. Given that leakage accounts for a sizeable portion of the overall power dissipation of VLSI circuits, it is also a crucial factor in power optimization. The main aim of this research is to shed more detail on the advances and breakthroughs in the field of deep submicron CMOS circuit power optimization.

The value of information is increasing in personal and professional environments. This results in duplicate files being stored in a storage area that can be misused. The aim of this proposed system is to maximize storage space. Improve storage and reduce file size with hash-based deduplication. The system allows the system to search and extract data using cryptographic hashing techniques such as SHA-256 to create unique identifiers of the data content. The program searches the archive system, calculates hash results for each archive, and finds duplicates based on matching results. Backup files are deleted or replaced with hard links to old files to save disc space. Backup files can be deleted or replaced with hard links to old files to save disc space. This work also addresses issues such as hashing and processing large files to improve performance. This technology facilitates local development because it provides more storage space without compromising data integrity. This technology is popular in computer and enterprise-level management because it provides a way to reduce data duplication and improve storage.

This paper provides an overview of underwater wireless communication modems, mainly focusing on optical, acoustic, electromagnetic and hybrid techniques. Underwater optical modems uses a light signal for transmission data, which offers high data rates but are limited by the absorption and scattering in water, hence, it is suitable only for short-range and clear-water applications. While, the acoustic modems supports for longer distances, but it face challenges like low bandwidth and high latency due to the characteristics of sound propagation in water. Whereas the hybrid modems are aimed to achieve the strengths of both optical and acoustic techniques, to achieve high transmission range and data rate by dynamically switching between the modems. This paper presents the design challenges such as energy efficiency, miniaturization and environmental adaptability. In addition to advancements improved in recent modulation techniques like adaptive algorithms and multi-modal communication protocols. This study identifies critical obstacles, including power constraints, environmental variability, and limited range in hybrid systems, while proposing potential solutions to enhance underwater communication for applications like oceanographic research, military operations, and environmental monitoring.

This paper presents a memory-efficient architecture for the fast Fourier transform (FFT), aimed at significantly reducing memory consumption and multiplexer usage. This design is particularly effective for high-throughput and low-latency applications such as real-time signal processing, wireless communication, and radar systems. Using four parallel memory modules, each storing N/4 samples, the architecture allows simultaneous read and write operations across all modules, thus minimizing latency and enhancing throughput. The design employs a consistent permutation scheme based on the perfect shuffle method, eliminating the need for re-configuring shuffling circuits and simplifying the design.Additionally, uniform read and write addresses across all memory units streamline control and facilitate memory consolidation. The architecture has been effectively deployed on a field-programmable gate array (FPGA). The proposed architecture realizes a 29% reduction in the utilization of look-up tables, alongside a 10% diminution in power consumption.Demonstrating its practical viability in systems that require optimized resource management and high performance.

This work presents a comprehensive evaluation of multiple word embedding models for natural language processing (NLP) applications. We begin by examining popular models and discussing desirable properties for both representations and evaluation methods. Our focus then shifts to analyzing six prominent models (BOW, TF-IDF, Word2Vec, fastText, GloVe, ELMo) through various intrinsic evaluators independent of specific downstream tasks. The results reveal that distinct evaluators capture various aspects of model quality, and some exhibit stronger correlations with real-world NLP performance. Additionally, we investigate the consistency of these intrinsic evaluators, offering insights into their complementary roles in assessing model effectiveness. This work provides valuable guidance for NLP practitioners, enabling them to select and refine word embedding models based on their specific task requirements and desired linguistic properties. This contributes to the development of more accurate and effective NLP applications across various domains.

Driver drowsiness is a critical factor contributing to road accidents worldwide. The accidents can be prevented if warning is provided on time to the drowsy driver. This research proposes a deep learning-based approach for detection of drowsiness in drivers. The system leverages computer vision techniques, specifically the calculation of Eye Aspect Ratio (EAR), to monitor and detect signs of drowsiness by analyzing facial landmarks. EAR is computed to determine eye closure patterns indicative of drowsiness. To enhance accuracy, a method for detecting sunglasses is integrated in the proposed system by utilizing Hue, Saturation and Value (HSV) color space and color masks to identify regions obstructing the eyes. Upon detecting drowsiness, the system triggers visual and auditory alerts to notify the driver, promoting timely intervention for safer driving. Experimental results demonstrate the system’s effectiveness in various lighting conditions and validate its ability to differentiate between drowsy and alert states with high accuracy. The proposed approach not only addresses the challenge of detecting drowsiness in diverse conditions but also contributes to mitigating road accidents caused by driver fatigue, thereby enhancing overall road safety.

A host of the Adaptable Security Information Systems (ASIS) utilizes Virtual LAN (VLAN), routing, security, and networking functions to address at par with organizational security requirements. The modular design makes it possible to adapt for its use by performing the necessary changes of falling under current security policies. It helps protect your data better; this can also saturate traffic flow with dynamic routing. Firewalls and access controls are advanced security features that help secure against threats. Port access controls provide further authentication mechanisms as well which contain risks for unauthorized access. By giving organizations using ASIS more flexibility and freedom of action, they also serve to enhance their ability for proactive response in the security world helping them be proactive which inherently builds resilience into a dynamic threat environment. The main Idea is to implement a Network security model on one of the known network simulators.

Parkinson’s disease (PD) is a progressive neuro degenerative condition that impacts millions globally. Early diagnosis is essential for improving treatment outcomes and enabling timely medical interventions. PD is primarily recognized by motoric symptoms, including tremors, muscles stiffness, and slow movements (bradykinesia). These motor impairments are often accompanied by changes in voice frequency and dopamine deficiencies. This project aims to employ a deep learning (DL) method for detection of Parkinson’s disease through both DaTScan imaging and voice analysis. A Convolutional Neural Network (CNN)-based VGG19 model is developed for detecting PD through DaTScan images, while an Artificial Neural Network (ANN) model is trained to analyze voice data to predict PD. This combined approach will serve as a robust model for PD diagnosis, allowing for a comprehensive and reliable detection method.

The brushless doubly fed reluctance machine (BDFRM) is a cageless, brushless, and has no winding on its rotor; it has been widely studied in a number of research projects. When compared to its induction equivalent, the brushless doubly fed induction machine (BDFIM), this feature gives the machine a number of advantages. The key benefits of BDFRM over BDFIM include lower maintenance, lower power losses, increased reliability, and robust in nature. The development of this machine technology has been limited by the complexity of its reluctance rotor design and the flux patterns for the indirect connection between the two windings located on the stator, namely the power winding and the control winding. Due to its straightforward design, easier in construction, and lower controller costs, the brushless doubly fed reluctance machine (BDFRM) has been explored as a potential replacement for the conventional doubly fed induction machine (DFIM) in wind turbine systems. Since the theoretical results have demonstrated that this machine functions effectively if it is well designed, it is evident that the BDFRM still continues to develop.

Alzheimer’s disease (AD) is a major public health issue that mainly affects older adults, resulting in progressive memory loss and various cognitive difficulties over time. Recent studies emphasize the necessity of diagnosing Alzheimer’s disease at an early stage. This disorder causes a steady reduction in cognitive processes, resulting in mental degeneration. In recent years, there has been a significant increase in programs targeted at detecting and preventing the progression of Alzheimer’s. According to research, genetics, stress, and diet all have significant factors in the development of this illness. Our study combines Convolutional Neural Networks (CNN) with Vision Transformer to diagnose Alzheimer’s disease using deep learning approaches. The CNN analyzes characteristics in brain images, allowing the system to differentiate between normal and AD-affected ones. We assessed the efficacy of our approach using the OASIS dataset (Open Access Series of Imaging Studies), providing valuable insights into its potential for early AD detection.The conclusion of our study outlines potential opportunities for future research and offers suggestions for upcoming investigations into the diagnosis of Alzheimer’s disease.

A method is proposed which avoids many limitations associated with traditional line loss estimation calculation. AI using neural network proposes the method of soft computing which gives accuracy and uniqueness both in linear and nonlinear operations. Line loss estimation incorporates variation of active and reactive load demand control treatments. In this paper, a 30-node system is tested where the system loss estimation is verified before and after training the data. The response was evaluated with change of total active and reactive power requirement by variations at different operating load conditions.

This research presents an innovative multimodal approach for diagnosing Wilson’s disease by integrating linguistic information with object detection methods, thereby enhancing interpretability and accuracy in MRI-based diagnostics. Utilizing deep learning models, specifically transformers for object detection, we incorporate language-driven analysis to detect disease-specific biomarkers more effectively in brain MRI scans. This approach leverages linguistic cues from medical images, allowing the model to better align visual information with diagnostic terms relevant to Wilson’s disease. Additionally, by incorporating voice synthesis alongside visual and textual modalities, our framework offers a richer, multimodal diagnostic tool that improves model confidence and accuracy, surpassing single-modality detection methods. This comprehensive setup underscores the potential of multimodal AI in supporting accurate and context-aware clinical decision-making.

Heart disease is one of the leading causes of mortality worldwide, and early detection plays a crucial role in preventing life-threatening complications. This research presents a robust heart disease prediction system that employs a support vector machine (SVM) classifier integrated with deep learning techniques. The system utilizes a hybrid approach to enhance the accuracy of predictions by combining the strengths of SVM’s decision boundary classification with deep learning's capability to automatically extract features from raw data. A large dataset containing various patient records, including clinical parameters such as blood pressure, and heart rate, is used to train the model. By applying feature extraction and dimensionality reduction techniques, the system reduces computational complexity while retaining critical information. The SVM classifier is fine-tuned using grid search and cross-validation methods to optimize performance. The model is evaluated based on its accuracy, outperforming traditional models. The integration of deep learning further enhances feature recognition and improves prediction robustness, making the system highly reliable for clinical applications. This approach is a scalable and efficient solution for heart disease prediction, assisting healthcare professionals in early diagnosis and decision-making.

A steady and reliable energy supply is essential to modern society since any disruption in the power supply can result in major operational, security, and financial issues. To improve fault detection in electrical grids and mitigate the impact of outages, this research suggests a hybrid machine-learning technique. The model incorporates the Random Forest (RF) and Support Vector Machine (SVM) algorithms, each with complementary benefits to increase the detection system’s accuracy and robustness. SVM excels in high-dimensional classification, effectively distinguishing grid stability states, whereas Random Forest provides ensemble-based robustness, reducing overfitting and enhancing prediction accuracy. When applied in MATLAB Simulink for real-time data simulation, the hybrid model achieved a high accuracy of 96% and showed balanced precision, recall, and F1 scores across stable and unstable classifications. This integrated approach is scalable and economically feasible for various grid scenarios due to its ability to reduce false positives, lower maintenance costs, and enable proactive anomaly discovery. The model has the potential to be a dependable solution for ensuring a consistent and uninterrupted power supply, which is essential for both security and societal advancement, as seen by its remarkable ability to detect flaws.

One of the most destructive natural catastrophes is flood, which seriously effects houses, infrastructure, and agricultural land, resulting in large financial losses and negative social effects. Ground surveys and hand inspections are too time-consuming, labour-intensive, and human error-prone traditional methods for evaluating postflood damage. Recent developments in machine learning and computer vision have created new avenues for automated, precise, and effective flood damage assessment. This research employs an ensemble deep learning framework that makes use of eight cutting-edge CNN architectures, such as VGG16, ResNet-50, and MobileNetV2. The objective is to use deep ensemble learning techniques to categorize flood detection. Our models were trained, tested, and validated using the FloodNet and flood area segmentation datasets. With a training accuracy of 99.8% and a test accuracy of 95.4%, the ensemble model performs better during the testing phase than a number of separate benchmark models. The suggested approach seeks to effectively forecast floods and carry out early evaluations of impacted regions. The suggested computer vision-based system seeks to provide near-real-time data on flood effects to help government organizations, disaster response teams, and insurance firms make educated judgments. This method offers a scalable way to improve postflood recovery operations by accelerating and improving the accuracy of damage assessment.

Precise crop classification is crucial for effective agricultural monitoring while current remote-sensing-based crop classification models leading to poor performance across different regions. This research focuses on the classification of crops in Agiripalli, a rural area on the outskirts of Vijayawada in Andhra Pradesh, using advanced remote sensing techniques utilized ML and DL models. Agriculture in this region is marked by diverse crop types and complex planting patterns, which present significant challenges for traditional crop classification methods. By leveraging Sentinel-1 and 2 satellite data through the GEE platform, we aim to develop a robust approach that enhances the precision of crop mapping and supports efficient agricultural resource management. Our methodology involves the application of both random forest (RF) and deep neural networks (DNN) to address the limitations of conventional methods. The random forest model demonstrates exceptional performance within the Google Earth Engine, achieving an overall accuracy of 98%. It also yields a macro-Average precision of 0.86, recall of 0.72, and an F1-score of 0.77, while the weighted average precision is 0.82, recall is 0.80, and the F1-score is 0.79. In parallel, the deep neural network model provides high precision scores, ranging from 0.91 to 0.95, with recall values between 0.50 and 0.94 and F1-scores from 0.65 to 0.81 across multiple crop classes. These outcomes highlight the advantages of integrating both optical and radar data to achieve superior classification accuracy, even under challenging environmental conditions. By optimizing crop identification and enhancing decision-making processes, this research contributes to more sustainable agricultural practices in Agiripalli, enabling better resource allocation, reduced wastage, and overall improved productivity.

This article proposes a dual-tuned, multifunction device using graphene and vanadium dioxide (VO2)-based metasurface for terahertz applications. It provides the capability of voltage tuning and temperature tuning to switch dynamically among the functionalities. The fundamental building block of the metasurface is designed using a lossy silicon dioxide (SiO2) substrate with a reflective gold-based ground layer at the bottom and an elliptical-shaped metasurface layer at the top. The metasurface layer is made of a combination of vanadium dioxide and graphene. At normal room temperature (298 K), the device can be employed as a triple-band absorber with sharp absorption points at 1.98 THz, 2.48 THz, and 3.11 THz with absorption of 96.37%, 99.63%, and 81.23%, respectively. At a higher temperature of 68 °C (341 K) or more, it can be employed as a linear-to-linear cross polarization converter (LTLPC) from frequency 1.62 THz to 2.96 THz, i.e., 58.52% fractional bandwidth (FBW) considering 90% polarization conversion ratio (PCR) level as well as a dual-band absorber with sharp absorption points at 1.69 THz and 3.05 THz with 90.89% and 98.54% absorption, respectively. The operating frequency of the device can also be tuned over the terahertz gap because of the voltage-tuning property of graphene. The response of the device remains outstanding up to the 50-degree opacity angle.

Machine learning, a significant part of AI, plays a vital role in daily life, supporting tasks like decision-making and real-world interactions. This technology’s ability to analyze data and make predictions has value across fields such as agriculture, healthcare, and finance. In this research, we address a key healthcare challenge heart disease, which has become a prevalent health threat, often without symptoms. To tackle this, we propose using machine learning to forecast heart disease risk by analyzing medical data, including age, blood pressure, and cholesterol levels. Early identification of high-risk individuals through predictive modeling can facilitate preventive healthcare and potentially lower heart disease incidence. The proposed model includes data processing, feature selection, training, and accuracy evaluation using precision, recall, and F1-Score as metrics.

The literature review focuses on optimizing traffic flow for traffic signal systems. The authors have traced the evolution of traffic signal control systems from the 12th century, both in theory and in real-world applications. They have also explored how technological advancements have been utilized to enhance traffic systems with limited resources. Over the last twenty years, the widespread use of AI and ML has significantly advanced the prediction and control of traffic signals.

Through the analysis of electrocardiogram (ECG) images, this study explores the application of machine learning and deep learning techniques for the early identification of cardiovascular illnesses. We evaluate the efficacy of conventional algorithms and cutting-edge deep learning techniques using a variety of datasets, preprocessing techniques, and feature extraction approaches model as the LSTM. While robustness is ensured by validation on separate datasets, the application of transfer learning improves model generalization. A focus on interpretability seeks to promote clinical acceptance, which is essential for incorporating new techniques into standard medical procedures. Overall, this study shows how these methods have the potential to transform the diagnosis of cardiovascular disease, allowing for prompt interventions and ultimately leading to better patient outcomes. Additionally, by permitting preemptive therapies, this interdisciplinary approach holds promise for revolutionizing cardiovascular healthcare.

Mangoes are valued both nutritionally and economically. However, Mango Leaf Disease presents a significant challenge for agricultural practitioners. Effective disease identification is crucial for addressing this issue, which requires extensive knowledge of plant diseases. Recent advancements in deep learning have significantly enhanced the ability to identify and classify plant diseases. The detection of illnesses in mango leaves using several machine learning and deep learning algorithms is covered in this research. The objective is to precisely detect and classify various diseases affecting mango leaves. The dataset is evaluated using several ML algorithms, including SVM, Decision Trees, Random Forest, Gradient Boosting, and KNN, as well as deep learning models such as CNN with sequential models, VGG16, VGG19, and YOLO. The proposed system leverages the latest versions of YOLO, specifically YOLOv8, YOLOv9, and YOLOv10 for the identification of mango leaf diseases. The models are trained on a custom dataset comprising eight classes. Among the various evaluated, the YOLOv8 and YOLOv9 models demonstrated significantly superior performance compared to other models with an accuracy of 98.83% and 99.16% respectively, highlighting their effectiveness in mango leaf disease detection. The model is designed to enable users to detect and classify diseases accurately without requiring expert intervention.

This paper combines a holistic approach with advanced deep learning techniques and Optical Character Recognition (OCR) to recognize symptoms from images of plants, thereby achieving the prospects of embracing plant diseases. It is undeniable that plant diseases are an important threat to agricultural productivity and require early detection and proper management practice. This article uses Convolutional Neural Networks and the ResNet architecture to diagnose plant diseases. The CNN model provides a multi-layer setup that does convolution, pooling, and finally makes it fully connected to categorize the images belonging to plants under specific categories of plant diseases using a dataset of plant photographs. A lightweight version of Inception-ResNet is also a combination of the power of residual learning along with CNNs in order to achieve enhanced classification performance. Each of the images contains information regarding the type of plant and its corresponding health or disease condition. Two models were well trained and tested on a comprehensive dataset that covered various plant species.Results have been found with high accuracy in classification, thereby The performance evaluation involves predicting diseases like tomato leaf mold, potato blight, black rot, and apple scab. With advanced deep learning architectures and preprocessing techniques, the models provide significantly high accuracy for real-time disease prediction in this approach. This can lead to better crop protection and agriculturally developed quality by possibly significantly contributing to fighting crop diseases for proper productivity development with precise and timely disease identification.

Classification of brain tumors is the most important thing in the diagnosis process with treatment of brain tumors, playing the biggest role in the treatment process as well as the prognosis of patients. Brain tumors could be benign or malignant and show extensive variability in their morphological as well as biological characteristics; they require a precise and timely diagnosis. Traditional diagnostic methods are mainly based on histopathological studies, which although they are definitive, are quite invasive and time-consuming. Recent developments in medical imaging technologies, including Positron Emission Tomography (PET), Computed Tomography (CT), and Magnetic Resonance Imaging (MRI), have completely transformed the field by providing real-time, noninvasive evaluation of tumor features. Here, image fusion plays a very important role in analyzing the image characteristics. This paper included some fusion methods and classification algorithms. It covers the various classification algorithms and fusion methodologies. This research focused on brain tumor classification based on MRI images. Intermediate fusion techniques, particularly those employing deep learning, consistently offer the best performance.

A neurological disorder known as Parkinson's Disease (PD) affects 60% of the population over the age of 50. Individuals with Parkinson's encounter movement impairments and communication obstacles, rendering physical appointments for treatment and monitoring problematic. Freezing of Gait (FOG) is extremely weakening yet inadequately comprehended sign of PD. FOG is an irregular gait pattern marked via incapability to initiate steps or turn when ambulating, especially in the sense of constricted environments. This syndrome compromises equilibrium, elevates the incidence of falls and diminishes quality of life. This FOGPD can be treated early, remotely, and correctly. Accurate diagnosis of PD necessitates robust Machine Learning (ML) and Deep Learning (DL) techniques together with effective medical instruments for evaluating neurological health. This research provides three machine learning strategies utilizing a hybrid model to detect partial discharge in its early phases. The results are contrasted with three applied ML techniques: Gradient Boosting, Decision Tree Random Forest along side a proposed Hybrid algorithm, which integrates Random Forest and Gradient Boosting.

Efficient resource allocation is critical for mobile adhoc networks due to their dynamic and decentralised nature. The integration of quantum computing techniques with the Vickrey Clarke Groves (VCG) auction mechanism offers a promising solution for optimising resource allocation. This paper proposes a Quantum enhanced VCG Auction (QVCG) framework that leverages quantum inspired algorithms to address computational complexity and enhance decision making precision. Experimental analysis demonstrates that QVCG achieves improved fairness, efficiency, and security in resource allocation, particularly in environments with high node mobility and constrained resources. The findings establish QVCG as a robust mechanism for supporting real-time and adaptive resource management in MANETs.

Artificial intelligence is transforming many industries, and its use in alternative dispute resolution holds revolutionary potential. ADR procedures such as negotiation, arbitration, and mediation are intended to settle conflicts outside the conventional court system and provide quicker more affordable and more adaptable resolutions. AI integrated into ADR can resolve the conflicts more quickly and effectively. AI system can enhance the decision-making for any complex matter in a simplified way and offer data-driven insights. The data that is analyzed through AI makes it possible to identify patterns which aids mediator and parties in forecasting results based on precedents and recommending the best solution. AI assists the arbitrators in documents analysis that speeds up the procedure for resolving and classifying large case files. By evaluating the possibility for settlement using data and providing guidance through machine learning algorithm which supports the arbitration and are likelihood to come to a conclusion through mutually agreeable terms. AI will alternative dissolution it is easy to generate a lengthy list for possible application or use cases in the ADR sector if one has a rudimentary understanding for artificial intelligence and large language model. But at the same time there are several drawbacks for using AI in ADR the problem that are related to the segment is thatthe there is issue related to the application usage and the ethical issue that crops up as a problem. Along with that the issue related to privacy and the possible detoriation for human judgment in conflict resolution. AI’s current use in ADR is additionally constrained by the absence for clear legal rules and standardization. This paper examines the ethical and legal ramification for AI- assisted dispute resolution, looks at the current role for AI in ADR and evaluates the possible advantage and disadvantages. It also examines the potential application for AI in ADR stressing the need for well- rounded strategy to optimize the benefits for the technology while resolving its drawback in ADR.

Protecting privacy of data is a critical issue when handling sensitive medical information and homomorphic encryption (HE) emerges as a promising method facilitating computation on encrypted data. However, securely and efficiently processing private information in cloud computing remains challenging. Fully Homomorphic Encryption (FHE) can be a possible solution to privacy issues such that untrusted third parties may process cipher data without compromising the confidentiality of information. FHE shall prove valuable in distributed computation environments where confidentiality and integrity of data is also equally important. This survey provides a comprehensive review of FHE on theoretical foundations, current developments, limitations, potential applications, and available tools. The survey also suggests that FHE can be combined with machine learning to enable efficient predictions while safeguarding patient data privacy.

VisionGrip has developed a revolutionary solution for individuals afflicted with Carpal Tunnel Syndrome (CTS) and Radial Nerve Palsy (RNP) through the introduction of a cutting-edge servo motor controller aimed at improving motor function and aiding in daily tasks. By harnessing electrooculography (EOG) signals, VisionGrip presents a non-intrusive technique for users to manage servo motors using natural eye movements. The sophisticated interface of the system interprets EOG signals in real-time, converting them into precise motor actions, thus empowering users to perform activities that demand delicate motor skills without depending on manual dexterity. This pioneering technology holds the potential to significantly enhance the quality of life for individuals grappling with CTS and RNP, providing them with a convenient and effective method of engaging with their surroundings while fostering self-reliance and independence. VisionGrip’s cost-effectiveness and ease of use further bolster its potential for widespread acceptance, positioning it as a promising tool for enriching the daily experiences of individuals coping with motor impairments.

The exponential growth of marine litter calls for a state-of-the-art detection and collection solutions in real-time. This chapter presents the enhanced underwater trash detection using the YOLOv11 network and ESRGAN. It provides a method for enhancing underwater images degraded with blurring and low visibility, along with refraction of light, thus improving visibility of objects within such water, which will enable YOLOv11 to detect marine debris quickly. With superior precision, recall, and mAP, the model excels in detecting small and partially occluded objects beyond what the current systems can do. This dual-pronged effort upgrades environmental monitoring, supports focused cleanup operations, and promotes marine conservation efforts. The study provides a strong framework in addressing underwater pollution, which is increasingly becoming a threat to environmental sustainability and marine biodiversity.

Traffic signal control (TSC) is an effective method for easing traffic congestion and enhancing traffic efficiency in urban areas, particularly with growing urbanization. Conventional traffic signal control techniques are unable to adapt swiftly to the intricate and dynamic road environment, necessitating a more intelligent approach to signal control. In recent years, there has been an increasing interest in employing reinforcement learning (RL) for TSC, which has displayed considerable potential in optimizing control strategies for complex traffic conditions. This work focuses on the application of RL in TSC for research purposes. To recreate real urban traffic conditions, a traffic simulation system was used in this study. Extensive experiments and comparative studies conducted in a simulated environment have illustrated the effectiveness and superiority of reinforcement learning methods in traffic signal control. This work primarily employs Deep Q-Network (DQN) in reinforcement learning to build the model and examines the variations in traffic signal timing under different reward mechanisms. The research findings demonstrate that reinforcement learning can significantly impact traffic signal control, effectively resolving current traffic signal control challenges and advancing the development of urban road network traffic efficiency.

In smart farming, IoT-related technologies are gaining popularity. The ultimate goal is to collect, monitor, and use important data for agricultural operations to optimize and sustain agriculture. Smart networks have advanced thanks to the Internet of Things. This research provides a LoRa-IoT-based efficient data prediction model for agricultural applications utilizing a Deep Maxout Neural Network (DMNN) and IoT Agriculture Dataset. LoRa is a leading Internet of Things technology because it can transmit vast distances with low power. A customized smart farming system using IoT and LoRa technologies and a low-cost, low-power, wide-range wireless sensor network is presented in this research. Data is analyzed using the DMNN architecture to estimate crop yields, irrigation needs, and disease outbreaks. Extensive trials show that the suggested approach outperforms existing predictive models in data processing and prediction. This model has 99.23% accuracy, 25.53% RMSE, 13.52% MAE, 9.62%, MAPE, 0.9905 R-square, and 1.49 s processing time.

The expeditious advancement of artificial intelligence (AI) facilitates a wide range of applications across different domains; however, it also introduces substantial challenges concerning speed and energy consumption, driven by the explosive growth of data. Spanning the breadth of research and real implementation, the rise of artificial intelligence (AI) and machine learning (ML) has fundamentally altered traditional critical thinking methodologies. This paper investigates the essential parts of artificial intelligence (AI) and machine learning (ML), emphasizing their crucial role in advancing the capabilities of optics and photonics technologies. Starting with basic definitions and frameworks, discuss different AI/ML algorithms used in optics.

This study explores the data mining techniques adopted by data scientists to solve analytical problems and examines their perceptions and strategies for choosing appropriate methods. In the era of big data, advanced analytics are essential for extracting valuable insights. With a wide range of data mining techniques available, selecting the right one can be challenging, particularly for inexperienced scientists. Using the technology acceptance model (TAM) and an exploratory qualitative approach, this research aims to provide guidance on technique selection. Data were collected through open-ended interviews and analyzed thematically, revealing key strategies and perceptions in data mining adoption.

This qualitative research examined public awareness of cryptocurrency scams through narrative analysis, identifying key knowledge gaps that contribute to victimization. Despite the increasing use of cryptocurrencies, many remain vulnerable due to limited understanding of the associated risks. The study, involving interviews with 10 participants from diverse backgrounds, explored the rise in scams, prevalent types, and the need for improved public education. Findings underscore the urgency for effective strategies to enhance awareness and protect investors, providing insights for developing educational tools and policies to prevent scams and improve investment safety.

Diabetes prediction is crucial for improving public health by enabling early detection and intervention. The integration of IoT in healthcare allows continuous monitoring of vital health indicators, such as glucose levels and BMI, through connected devices. This real-time data, combined with machine learning, enables more accurate and personalized interventions. In this study, we propose a Bagging Classifier model enhanced with IoT-driven data, achieving 99% accuracy. Trained on a balanced diabetes dataset from Kaggle, the model outperforms traditional classifiers, demonstrating its potential for diabetes prediction and management in IoT-enabled healthcare systems.

We propose a method for identifying and classifying skin diseases using the HAM10000 dataset, which contains approximately 10,015 images from ISIC. The approach combines the ResNet-50 neural network for feature extraction with enhanced random forest (ERF) for classification. Compared to traditional algorithms, our model accurately identifies lesion edges and improves classification reliability. The proposed method achieved 95% accuracy on the HAM10000 dataset, outperforming existing algorithms and demonstrating its potential for reliable, automated skin disease detection.

This research develops an IoT-enabled predictive model for early stroke detection and continuous patient monitoring, combining XGBoost with IoT devices to collect and analyze health data in real time. The model integrates machine learning techniques, including feature selection and regularization, achieving high accuracy in stroke prediction. Evaluated using a Kaggle dataset, it demonstrates improved sensitivity for early detection based on risk factors like age, blood pressure, and lifestyle. These findings highlight the potential of IoT and machine learning to transform stroke care by enabling proactive intervention and reducing healthcare system burdens.

This work presents the importance of pick-and-place robot and its wide range of uses in the present world, this paper tell us how efficiently the robot is being used in the current world. It is operated with the help of UNO and an HC-05 Bluetooth module for control. It has an immersive accuracy, allowing it to be controlled for versatile operations by manually or with the help of any application. This system uses three servo motors: one to move the hand in the direction of the object, the second (2nd) one for the height adjustment of the hand, and lastly the third (3rd) one for grabbing the object. It is versatile enough for use in logistics and small-scale manufacturing, providing a low-cost and scalable and reducing labor charges. It is reliable and efficient, successfully completing tasks like picking up, moving, and placing objects with accurate and precise system.

Diabetes is a problematic condition all over the world, and it has huge impact on global population and healthcare budgets. Diabetes is a long-term illness that impairs the body’s ability to properly regulate blood sugar, and the search for efficient methods of early diagnosis is still relevant now. This project is the proposal of artificial neural network in examining diabetic disease using clinical characteristics. The proposed system utilizes a low cost and effective technique for data analysis in context with medical data-driven diagnosis for possibility of developing diabetes. The framework requires clinical aspects including blood glucose level, blood pressure, BMI, insulin levels and family medical history for the assessment of tested models. That way, the system offers high accuracy and reliability—thanks to algorithms like gradient boosting classifier, SVM and KNN. The application of feature selection approaches ensures that only useful indicator domains are chosen, to minimize computational costs as well as improve model interpretability. The gradient boosting algorithm significantly improved the results in our diabetes detection, showcasing the effectiveness of predictive modelling in health care. The accuracy score received by gradient boosting algorithm is 91.45, which is best compared to other algorithms. Based on this analysis, ML does have the capability to revolutionize diabetes diagnosis which in return offers wider and more accurate proactive healthcare systems. Tentative developments include incorporating real-time monitoring devices into the models in the future and expanding their applications to other chronic diseases, which can result in holistic and sustainable healthcare programs.

This paper aims to introduce a deep learning-based architecture for hand gesture classification and introduce a potential pathway for the improvement of Brain Computing Interfaces (BCI)-fostered motor rehabilitation systems. The proposed mechanism included inspecting the sequential and ensemble Electromyography (EMG) denoising strategies to remove potential artifacts from the EMG signals, both the strategy included working with Dynamic Mode Decomposition (DMD), Variational Mode Decomposition (VMD), and Tunable Q-factor wavelet transform (TQWT). The refined EMG signals are fed in the custom-designed Convolutional Neural Network (CNN)–Bidirectional Long Short-Term Memory (BiLSTM) based deep neural network architecture suited for the classification task. The research task is performed in three steps, the first step includes training the deep learning model without any artifact rejection, the second step comprises training the model with sequential EMG denoising, and the third step includes training the model with ensemble EMG denoising. The results depicted commendable performance with the ensemble denoising approach reaching an accuracy of 99%, thereby curving a potential pathway for the enhancement of BCI-fostered motor rehabilitation systems.

Deep learning technologies show an impressive performance in the machine learning community for biomedical image detection, segmentation, and classification. They become the widely used computational approach to achieve complex tasks matching to humans. Understanding the applications, algorithms, and challenges is crucial for the development of DL-based architectures. Massive data learning is one of the main advantages of deep learning. In this paper, we provide a comprehensive review of deep CNN architectures—LeNet-5, AlexNet, VGGNet, GoogLeNet, and ResNet—highlighting elements of the particular model along with architectures. Improvements of the one model over the previous model were also discussed. Though they are classic and old-fashioned architectures, when they integrate with one or more other CNN models, they yield significant results in the field of disease detection and classification. This paper outlines the current work that carried out by these models in biomedical field recently and finally concludes with a comparative summary of all five models that are discussed.

Food waste is a global issue that has a significant impact on the environment and economy of nations. It is aware that the causes of food waste are complex and varied, arising from several points in the food chain, including production, distribution, retail, and consumption, particularly in the bakery sector. Furthermore, because bread has a limited shelf life and goes bad after only three days, a lot of it is wasted. Model prediction is believed to be one of the ways to tackle this issue as by predicting the sales in the future, managers will be able to manage the daily production to reduce wastage. In this paper, sales data from a French bakery store is gathered, and SARIMA, XGBoost, and Random Forest is implemented to generate an accurate model to predict the sales in the upcoming year. This is motivated by the need to reduce the amount of bread wasted every day, month, or year. The result shows that XGBoost outperforms both SARIMA and Random Forest with the lowest RMSE value at 123.61. The research results can help retailers and bakeries reduce the amount of bread wastage globally through better inventory management as well as bread production to improve both profitability as well as the environment.

Breast cancer is a major global health issue for women. Timely detection and precise classification are important in effective treatment and enhancing chances of survival. Historically, experts have used mammography and MRI in classification, which depend on expert interpretation and vary in accuracy. This study investigates the use of deep learning models, VGG16 and EfficientNet B7 in particular, to classify breast cancer images. These models focus on three classes: benign, malignant, and normal based on Breast Ultrasound Images Dataset (BUSI). The report identifies how these models revolutionize medical diagnostics. It analyzes the tremendous significance of transfer learning along with fine-tuning toward improving accuracy while reducing overfitting tendencies. The findings from this research indicate that model selection must be specific to medical imaging tasks.

Cloud computing has emerged as a prominent area of research, reflecting the widespread adoption of technology across various facets of contemporary life. However, alongside its benefits, cloud computing presents significant challenges in ensuring the privacy and security of user data. The propensity for users to upload unsecured data to cloud platforms renders it susceptible to unauthorized access and breaches. In response, this chapter introduces a robust solution aimed at safeguarding on-cloud data from potential threats. Through the implementation of the proposed system delineated herein, the aim is to reduce the risk possibilities of an unauthorized access, thereby bolstering the privacy and confidentiality of user-uploaded data. By addressing these concerns, this research seeks to enhance the trust and reliability of cloud computing storage services while fostering a safer digital environment for users and organizations alike.

The job openings for software engineers are growing significantly along with rapid technological development. Recruiting software engineers in an effective and efficient manner has become a crucial topic in the business domain. Making recruitment decisions is a complex process, as multiple aspects define a good-fit candidate. The hiring personnel are required to gather and act based on complex, multi-dimensional data. The currently proposed solutions are lacking in the concern of approximating technical skills from multiple input variables. To overcome the problem, this research proposes a fuzzy logic-based decision support system that aims to assist hiring personnel in making hiring decisions for software engineers. It utilizes hierarchical architecture and introduces a modification to the Mamdani model for a more intuitive and logical output.

In the face of Malaysia’s digital challenges, ubiquitous computing emerges as a promising avenue to enhance efficiency and bolster economic impact within the realm of public transportation. This research investigates the integration of ubiquitous computing technologies into Malaysia’s Mass Rapid Transit (MRT) and bus systems, aiming to optimize operations, improve service quality, and stimulate economic growth. Through a case study approach, this paper explores the potential benefits of ubiquitous computing in addressing key challenges faced by Malaysia’s public transportation sector. Leveraging real-time data analytics, predictive maintenance algorithms, demand prediction models, dynamic pricing mechanisms, and customer sentiment analysis, the proposed framework seeks to revolutionize how MRT and bus systems operate and interact with passengers. By enhancing efficiency and service reliability, ubiquitous computing integration promises to attract more ridership, reduce congestion, and ultimately contribute to a more sustainable and vibrant economy. Through this research, the researchers aim to shed light on the transformative potential of ubiquitous computing in reshaping Malaysia’s public transportation landscape, offering insights and recommendations for policymakers, transportation authorities, and stakeholders to harness the full economic benefits of digital innovation.

The aim of this study is to enhance the quality of life for individuals who stutter and to raise public awareness about stuttering. AFPWS is a comprehensive, user-friendly app designed to support and empower individuals who stutter. It is a platform that offers a variety resource to help users improve their speech fluency, build confidence, and communicate more effectively. The findings revealed that participants held stereotypes about people who stutter, as evidenced by the predominantly negative words they used to describe them. More than half of the participants, mostly male, prefer responses from humans over chatbots, even if it takes longer, likely due to negative experiences with chatbots not understanding their questions. Therefore, the implementation of natural language processing and machine learning are important to understand and generate appropriate outcome desired by the users. The chatbot is trained using sequential model and the data is analyzed through the implementation of data visualization.

Code smells, indicative of poor code quality and maintainability, pose significant challenges to software development teams. Despite the availability of automated tools, accurately identifying and resolving these issues remains a complex task. To address these limitations, this research addresses the shortcomings of existing code smell identification and resolution methods. By leveraging cloud-based crowdsourcing and interactive learning, the authors propose a novel platform that effectively identifies and rectifies inefficient code design choices that evade traditional compiler detection. The approach aims to overcome the limitations of automated tools, which often struggle to achieve high accuracy, by providing a semi-automated refactoring strategy that empowers developers to make informed decisions regarding recommended code modifications. Through a combination of quantitative and qualitative analyses, the research demonstrates how the proposed solution improves the efficiency and effectiveness of code smell identification and resolution by addressing key challenges such as a lack of knowledge, insufficient validation support, and limited collaborative opportunities.