Journal of Intelligent Systems and Internet of Things

Channel estimation poses critical challenges in millimeter-wave (mmWave) massive Multiple Input, Multiple Output (MIMO) communication models, particularly when dealing with a substantial number of antennas. Deep learning techniques have shown remarkable advancements in improving channel estimation accuracy and minimizing computational difficulty in 5G as well as the future generation of communications. The main intention of the suggested method is to use an optimal hybrid deep learning strategy to create a better channel estimation model. The proposed method, referred to as optimized D-LSTM, combines the power of a deep neural network (DNN) and long short-term memory (LSTM), and the optimization process involves the integration of the Reptile Search Algorithm (RSA) to enhance the performance of  deep learning model. The suggested hybrid deep learning method considers the correlation between the measurement matrix and the signal vectors that were received as input to predict the amplitude of the beam space channel. The newly proposed estimation model demonstrates remarkable superiority over traditional models in both Normalized Mean-Squared Error (NMSE) reduction and enhanced spectral efficiency. The spectral efficiency of the designed RSA-D-LSTM is 68.62%, 62.26%, 30.3%, and 19.77% higher than DOA, DHOA, HHO, and RSA. Therefore, the suggested system provides better channel estimation to improve its efficiency.

Brain tumor is an abnormal development of brain cells that, if left untreated, can have severe consequences. Brain tumour semantic segmentation is the process of determining and distinguishing the impacted brain regions, which is essential for accurate diagnosis, treatment planning, as well as surveillance of the tumor's development over time. This paper presents a model for identifying and segmenting brain tumor using Unet architecture with the optimization of hyper parameters using the Moth Flame Optimization (MFO) algorithm. Due to its capacity to collect spatial information, the Unit architecture is a common choice for picture segmentation tasks. The MFO algorithm is an optimization technique that draws inspiration and replicates from the behavior of moths. Both techniques are developed to improve efficiency. The performance of the model has increased using the MFO method, which led to improved segmentation results. Based on comparative analysis report, the proposed model shows a percentage improvement of approximately 65.16% in MSE, 28.87% in PSNR, and 40.30% in Tversky compared to the Unet and Unet++ models. This method has demonstrated good results in identifying and segmenting brain tumors, which can be helpful in the early identification and treatment of brain tumor.

Microscopic examination of tissues to detect oral cancer falls short as traditional microscopes struggle to easily differentiate between cancerous and non-cancerous cells. The identification of cancerous cells through microscopic biopsy images has the potential to alleviate concerns and improve outcomes if precise biological approaches are employed. However, relying solely on physical examinations and microscopic biopsy images for cancer identification increases the likelihood of human error and mistakes. Therefore, in order to obtain accurate results, a new research technique has been developed. In this manuscript, Gazelle Optimized Visual Geometry Group Network with Resnet101 fostered Oral Squamous Cell Carcinoma Detection (OCD-VGGNetCNN-GOA-Resnet101) is proposed. In this method initially, the images are taken from Kaggle repository benchmark dataset and preprocessed to improve image quality.  Then the result is given to the Visual Geometry group Network based CNN (VGGNetCNN) with Resnet101 for classification. Finally, the VGGNetCNN -ResNet 101 classifies image into normal and OSCC. Then the simulation performance of the proposed -VGGNetCNN-GOA-Resnet101 method attains 23.67%, 34.89%, 39.45% and 45.31% higher accuracy while compared with existing methods such as OCD-CNN-Alexnet, OCD-CNN-VGG19 and HI-OCD-CNN-INet respectively.

Vehicle tracking is one of computer vision's most important applications, with applications ranging from robotics and traffic monitoring to autonomous vehicle navigation and many more. Even with the significant advancements in recent research, issues like occlusion, fluctuating illumination, and fast motion still need to be addressed, calling for more investigation and creativity in this field. This study performs a thorough examination of various vehicle-tracking approaches and suggests a thorough classification scheme that divides them into four main categories: strategies that rely on features, segmentation, estimate, or learning. Two well-known methods are highlighted specifically in the estimation-based category: particle filters and Kalman filters.

The Machine Learning domain has made a major process with the progression of state-of-the-art technologies. Since current algorithms often don’t provide palatable learning performance, it is necessary to continually upgrade them. This paper has illustrated the comparison of the Long Short-Term Memory (LSTM) model and the Artificial Neural Networks (ANN) model in the prediction of the Internet Movie Database (IMDb) website. These evaluations were then related to sentiment assessment approaches to evaluate their predicted accuracy and performances. The results demonstrate that the ANN model outperforms the LSTM model with a high accuracy rate in terms of the prediction accuracy and loss indicators for the IMDb movie review’s sentiment analysis task in terms of the prediction accuracy and loss indicators for the IMDb movie review’s sentiment analysis task. The accuracy of prediction on the test dataset of the ANN model is 83.5 % and the LSTM model is 83.5%. Therefore, it can be concluded that the standard artificial neural network model that was utilized is an appropriate technique for sentiment assessment tasks in IMDb rating text data.

An influx of smart spaces that are now connected to the IoT network has increased new forms of cyber threats; thus, a need for more effective IDS to deal with these complex cyber threats. Traditional security measures cannot solve the modern problem of protecting IoT devices as they are a complex and homogeneously distributed network. Advancements and development of Artificial intelligent (AI) and machine learning technologies have provided new hope to make more reliable IDS. Our study presents Particle Swarm Optimization integrated Light-Weight Gradient Boosting Machine, abbreviated as LGBM-PSO in which, the PSO algorithm is applied for hyper parameters optimization in the model training. Based on the ensemble methodology, a new model for network intrusion detection is proposed in this study to improve the accuracy of the technique proposed. As for the current study project, the “DS2OS” dataset was employed to execute the suggested task. All of the data obtained from the traces of the smart devices placed in a smart home environment are incorporated in this dataset. The IDS model comprises several stages, one of which comprises data preprocessing that entails data cleaning, normalization, and encoding of network traffic data. Feature selection and dimensionality reduction are used which leads to the optimization of the dataset in this case. The core of the model comprises four classifiers: The compared models are Decision Tree (DT), LGBM-PSO, Light Gradient Boost Machine (LGBM), and Extreme Gradient Boost (XGB). Each of these classifiers can be combined with a majority voting ensemble method to increase the reliability of the predictions. The suggested model's accuracy that is LGBM-PSO is the highest with a value of 99.89%. The corresponding figures for the training data are 99.79%. Stand on the testing data proving the efficiency and stability of the algorithm. The use of the ensemble approach is superior especially when using a complex model like LGBM-PSO in the field of intrusion detection. As a result, high accuracy, optimized time, and effective threat identification ensure that it is a useful tool in strengthening security in the different applications.

This research paper explores the significant impacts of multiple loan fraud on Indian banks and financial institutions, emphasizing the resulting bad debts and financial losses. The issue is exacerbated in the real estate sector, where influential developers exploit system vulnerabilities to secure multiple loans using the same collateral. Consumers also face challenges in accessing credit due to these fraudulent practices. The study underscores the need for enhanced regulatory measures and internal controls within financial institutions. Additionally, it introduces IoTBlockFin, a decentralized system that integrates block chain and IoT technologies to securely assess customer reliability and mitigate fraud. IoTBlockFin's Advanced Proof of Work (APOW) mechanism, combined with IoT data for real-time monitoring, offers superior security, latency, and cost-effectiveness compared to centralized systems, as demonstrated by experimental results.

When it comes to hospital administration, patient care, and medical data analysis, the Healthcare Internet of Things (HIoT) is nothing short of a paradigm revolution. We dive into this new paradigm to examine its far-reaching effects and revolutionary possibilities in the healthcare system. The context is established by introducing HIoT as a game-changing development in healthcare. Using the IoT to network several devices, this model paves the way for real-time patient monitoring, streamlined inventory management, and integrated telemedicine. The healthcare industry as we know it will be transformed by HIoT as it strives to improve resource allocation, simplify operations, and provide proactive patient care. Our investigation includes a thorough appraisal of how HIoT will affect many facets of medical treatment. We use many research approaches and quality indicators for this evaluation. We may evaluate the viability and scalability of HIoT solutions by testing them in experimental settings that mimic real-world healthcare settings. To provide a precise depiction of the healthcare system, dataset environments use well maintained medical data sources. The performance and efficacy of HIoT technologies may be evaluated using measurable criteria such as sensitivity (0.94), specificity (0.89), F1-Score (0.91), ROC-AUC (0.95), and cost savings ($150,000). To determine the relative importance of each part of the HIoT ecosystem, researchers undertake "ablation studies. Our findings provide a clear picture of the disruptive potential of HIoT. Better patient outcomes may be ensured via early interventions thanks to the improved accuracy (0.92), efficiency (9.2), and satisfaction (9.2) provided by the suggested HIoT technique for patient monitoring. When healthcare and telemedicine are combined, the success rate of remote consultations increases to 95%, response times decrease to 15 minutes, and more people have access to medical treatment.

Face recognition technology is gaining popularity for security, access management, and user identification. A novel facial recognition method employing cutting-edge deep learning algorithms and attention processes reduces false positives in this study. This technique was designed to approach facial recognition differently. We demonstrate statistically substantial recognition gains over current approaches through extensive research and experimentation. The recommended solution uses an attention device and a complex feature extraction module. The pieces work together to highlight distinctive characteristics and facial identifiers. To optimize performance and generalization across datasets, data addition and hyper parameter adjustment are used to fine-tune the model. We do this for maximum benefit. Studies on the issue may help us understand the multiple reasons that make ablation so successful. We also discuss facial recognition technology's moral difficulties, including fairness and user privacy. We also emphasize cautious distribution. Our findings expand facial recognition technology knowledge and pave the way for future studies. This study demonstrates that better Mobile Net models and Internet of Things technologies increase the accuracy of mobile facial recognition. The project overcomes the challenge of providing powerful AI tools in resource-constrained situations by utilizing IoT infrastructures and effective, lightweight Mobile Net architectures. Extensive testing demonstrates that the technique increases identification rates and outperforms existing models, showing its suitability for real-time operations. The Internet of Things enables data mobility and cross-device model usage. This illustrates that the IoT ecosystem can enable effective and scalable security solutions.

Localization is an extremely important component of applications that make use of wireless sensor networks. It has a substantial impact on academics as well as real-time sensor deployment applications in the aim of lowering the amount of energy that is used while simultaneously locating unknown nodes. The process of obtaining the coordinates along an axis that represent the locations of the sensor nodes is referred to as localization. The accuracy of locating the positions of the nodes varies depending on the environmental conditions, the type of nodes, the type of application, and the type of localization methods used. A standard localization method known as distance vector hop (DV-hop) localization will be able to determine the positions of unknown nodes with typical accuracy with the assistance of beacon nodes based on Internet of things. The DV-hop and improved weighted centroid localization algorithms, in addition to the suggested boosted weight centroid-based localization approach, are both addressed in this article. The suggested boosted weight centroid localization technique is utilized to find nodes in the remote area of the WSN while conserving energy. This is accomplished with the assistance of measurements involving both the nodes and the centroid. The modified weight metric is utilized in the process of carrying out the task of localisation of an unknown node. The performance of BWCLA is evaluated based on a number of different metrics, including accuracy in localization, average localization error, total packets utilized, and energy usage.

Healthcare Internet of Things (IoT) initiatives that aim to integrate technology and medicine are shaking the sector to its foundations. The revolutionary potential of the proposed strategy is shown here as we investigate the far-reaching consequences of the Healthcare IoT on remote patient monitoring. The beginning sets the stage by underlining the significance of bridging the gap between technology and medicine. Our multi-pronged approach comprises Internet of Things (IoT) remote monitoring, cloud-based analysis, artificial intelligence (AI) integrated diagnostics, real-time alerts, and predictive analytics. Our study's results demonstrate that the proposed approach is superior to the status quo. The area of remote patient monitoring has profited considerably from the employment of traditional approaches, such as the fusion of data from wearable sensors, analysis in the cloud, diagnostics that utilize artificial intelligence, real-time monitoring, predictive modeling, and smart alarm systems. The suggested strategy, however, performs very well across all of the most important measures of assessment. Comparatively, the accuracy rate of the conventional wearable sensor fusion approach was only 76%, whereas our suggested method reached 89%. Our strategy was also more accurate than the standard approach (88% vs. 73%). When compared to the recall rate of 68% produced by conventional methods, our suggested strategy significantly outperformed the competition. It's a great option for hospitals and clinics since it improves diagnostic precision and speed without breaking the bank.

This research compares federated and centralized learning paradigms to discover the best machine learning privacy-model accuracy balance. Federated learning allows model training across devices or clients without data centralization. It's innovative distributed machine learning. Keeping data on individual devices reduces the hazards of centralized data storage, improving user privacy and security. However, centralized learning concentrates data on a server, which raises privacy and security problems. It evaluates two learning approaches using simulated data in a simple regression problem framework. Federated learning seems to be as accurate as centralized learning while protecting privacy. The paper also shows how federated learning works in popular machine learning frameworks like TensorFlow Federated. This research shows that federated learning protects privacy while producing accurate machine learning models. It challenges the idea that machine learning must constantly choose between privacy and accuracy. Empirical facts and theoretical ideas from this study advance machine learning methodology discussions. In the digital era, it promotes privacy-conscious, dispersed learning frameworks.

Efficient compression algorithms are required to handle the growing amount of medical picture data, ensuring that storage and transmission requirements are met without compromising diagnostic quality. This research presents a hybrid image compression framework that integrates deep learning alongside standard lossless compression techniques. A convolutional autoencoder (CAE) learns a compact representation of medical images, which are subsequently compressed using the Brotli algorithm. Our technique beats conventional approaches, like JPEG, JPEG2000, and wavelet-based ones, according to an analysis of a brain MRI dataset. It maintains competitive compression ratios while producing higher (PSNR) and (MSE), indicating higher picture integrity and low information loss. To strike a good balance between the critical need for accurate diagnosis and the economical use of resources, this study offers a possible method for compressing medical images.

The agricultural industry faces significant challenges in improving efficiency and productivity, particularly in monitoring crop health and environmental conditions. Traditional methods are often labor-intensive, time-consuming, and lack real-time data, leading to suboptimal decision-making. Recent advancements in Internet of Things (IoT) and Artificial Intelligence (AI) technologies offer promising solutions. Long Range (LoRa) communication, a type of low-power wide-area network (LPWAN), enables long-distance data transmission with minimal power consumption, making it ideal for rural and expansive agricultural areas. When combined with deep learning, which can analyze large volumes of data to generate predictive insights, these technologies have the potential to revolutionize agricultural practices by providing farmers with timely and accurate information to optimize crop management and resource utilization. This study introduces an intelligent mote for agricultural applications, leveraging Long Range (LoRa) communication and deep learning techniques to improve precision farming. Traditional agricultural monitoring methods are labor-intensive and lack real-time insights. To address this, the mote is equipped with sensors to monitor temperature, humidity, soil moisture, and light intensity, transmitting real-time data over long distances with minimal power consumption using LoRaWAN. The collected data is processed by deep learning models to predict crop yield and identify potential issues. Field tests demonstrated a 15% improvement in yield prediction accuracy and a 20% reduction in water usage compared to traditional methods. These results highlight the effectiveness of integrating LoRa and deep learning in enhancing agricultural resource management and productivity.

Today's clinical analysis and precise illness detection are mandated requirements for the development of intelligent expert systems. Since lung cancer affects both men and women equally and has a greater mortality rate than other illnesses, a more complete examination is needed to diagnose lung cancer. More helpful information regarding a lung cancer diagnosis may be provided by images from a computer tomography (CT) scan. Various machine learning and deep learning algorithms are created to enhance the medical treatment process using CT scan input pictures. But research still has a bad side when it comes to creating a precise and intelligent system. In order to improve the detection of lung tumors from the CT input images, this paper presented Firefly optimized pre trained transfer learning. The previously trained model VGG-16 is used in this paper to extract features more effectively, using the features chosen via the firefly optimization approach to increase classification accuracy while reducing complexity. The thorough testing done with the “LUNA-16 & LIDC Lung image” datasets is assessed & studied along with other performance measures like "accuracy, precision, recall, specificity, and F1-score". Investigation results show that the suggested design outperformed the “DenseNet, AlexNet, Resnets-50, Resnets-100, VGG-16 & Inception models” and reached the top results with "98.5% accuracy, 99.0% precision, 98.8% recall, with 99.1% F1-score.