Clouds, Internet of Things (IoT), Vehicular networks founded on the cutting edge of Information and Communication Technologies have redefined both data access and its use. But today that more and more is about an increasing number of bits traveling over networks - something that really heightens the chances for online attacks and bigger security breaches. To insure the integrity, confidentiality and availability of data have to give an assurance for the security in recognition of the intrusion intrusion detection systems have been incorporated into network security today. On integrating machine learning and deep learning algorithms into IDS design: A new perspective to improve the accuracy of the detection system to combat against evolving cyber threats. This work uses the more recent ML and DL algorithms to enhance the efficacy of IDS with respect to their accuracy and computational complexity. The experimental results also highlight the superiority of ensemble techniques and deep neural networks in intrusion detection. The best overall performance in our experiments presented the Random Forest 99.51 % accuracy and F1-score, followed by Gradient Boosting with 99.15 %, and SVM $\mathbf{9 9. 1 0 \%}$. Even neural network, with LSTM model achieved a comparable performance of $\mathbf{9 9. 0 7 \%}$. The work indeed illuminated the accuracy-computational trade-offs. The simpler algorithms like Decision Trees and KNN had quite quick training & prediction which in turn were more applicable for real time usage. Gradient Boosting and neural networks are more expensive to train but have better ability of detection. Results of this study indicated that an optimal accuracy and computational cost balance model will be essential in utilizing the ML and DL technologies with IDS, thus appreciably improving their pervasiveness to the end users towards eliminating cyber threats for real-time network security.

Abstract One hallmark of PD is the degeneration of dopamine-producing neurons in substantia nigra of the brain, and resulting in motor impairments, such as tremor or rigidity. It has been difficult to make an early and accurate diagnosis because of overlapping symptoms with other neurological diseases and limitations in conventional diagnostic tools. In order to tackle the challenges in PD, we propose a new hybrid architecture of SRGAN and SCO. Unlike previous approach for Alzheimer’s our method enhances low-resolution MRI and PET images presenting higher diagnostic potential. The approach could be carried out in clinics with limited screening capability. Super-Resolution GAN (SRGAN) An SRGAN was constructed which used a customised variation of DeepResNet (i.e. ResNet-150 with skip connections), the Generative Adversarial Network (GAN) and the Gaussian filter to enhance the quality of the MRI and PET images. Various operations were performed on the input images as pre-processing, such as, contrast stretching, noise filtering, gamma correction and pixel-normalization to clear input for the model. Our SRGAN model was better performing than the existing CNN models like DenseNet, AlexNet, VGG19 and ResNet with an accuracy of 96.5%, F1 score of 94.5% and balanced accuracy (BAAC) of 93.5%. Moreover, the skip connection optimization (SCO) of the SRGAN improved over various optimization methods such as GA, ACO, GWO, SGD, ADAM and AdaMax. The results demonstrate that the proposed model outperforms existing methods for the prediction of Parkinson’s disease.

Monocular depth estimation, the process of inferring 3D scene geometry from a single 2 D image, is a pivotal challenge in computer vision with applications spanning autonomous navigation, augmented reality, and 3D reconstruction. The task's inherent ill-posedness—stemming from the ambiguity where multiple 3D scenes can project onto the same 2D plane—demands sophisticated models capable of leveraging contextual cues and learned priors. We present FullHybridDepthNet, an innovative hybrid architecture that integrates Convolutional Neural Networks (CNNs) for local feature extraction, Transformers for global context modeling, and implicit neural representations for continuous depth refinement. Enhanced by a multi-objective loss function, gradient-weighted attention mechanisms, and contrastive learning, our model achieves superior performance on the NYU Depth V2 dataset, with accuracy of 0.878, RMSE of 0.412 meters, and inference time of 42 ms on an NVIDIA RTX 3090 GPU. This work pushes the boundaries of real-time, high-accuracy depth estimation, offering a robust and scalable framework for practical deployment.

Portfolio optimization remains a critical challenge in modern investment management, particularly as traditional methods like Modern Portfolio Theory (MPT) and Value at Risk (VaR) show limitations in addressing complex, multi-dimensional investment objectives. This paper presents a novel approach to portfolio optimization using genetic algorithms (GAs), designed to overcome the constraints of conventional optimization methods while accommodating diverse investor preferences and market conditions. Our methodology implements a flexible, multi-factor optimization framework that processes multiple input parameters, including but not limited to asset prices, volatility, correlations, sustainability ratings, and regional preferences. The system employs a fitness score calculation mechanism that aggregates weighted factors across three primary categories: risk/return metrics, portfolio constraints, and asset characteristics. Through extensive validation testing across 8,000 different test cases, our GA-based approach demonstrated an average 30% improvement in portfolio fitness scores compared to initial portfolios. Notable results include successful optimization of sustainability-focused portfolios, with targeted allocations achieving 60.4% median weighting for sustainable assets when specified. The framework's ability to handle multiple objectives while maintaining computational efficiency (8–10 second response time) represents a significant advancement in personalized portfolio management. Our findings (accuracy of 91%, AUC score of 92%, and Log loss of 0.27) suggest that genetic algorithms offer a more robust and adaptable approach to portfolio optimization compared to traditional methods, particularly in addressing the growing demand for personalized investment solutions and specialized investment criteria such as ESG considerations.

This paper introduces an efficient, unified system for product rating prediction and content-based personalized recommendations using Amazon data. With the surge in online shopping, accurately predicting user ratings and delivering personalized recommendations are essential for enhancing customer experiences and boosting sales. Our system leverages machine learning and natural language processing to analyze Amazon’s extensive datasets, including product reviews, descriptions, and user profiles. Our innovative approach, which concentrates on content-based methods to predict product ratings and generate recommendations, has been validated through experimental evaluations to significantly improve prediction accuracy and recommendation quality compared to traditional methods. This unified system provides a robust solution to the challenges of rating prediction and personalized recommendations in the e-commerce sector. Overall, the support vector machine (SVM) and random forest (RFT) performed well compared to other existing models. It delivers tangible benefits to consumers and sellers on the Amazon platform, fostering a more efficient and satisfying shopping experience.

The use of quantum-mechanical phenomena to solve problems that traditional computers cannot solve is known as quantum computing. Every day, we reap the benefits of traditional computing. How-ever, there are some issues that classical computers cannot handle, and we lack the processing capability to address them. This calls for the development of quantum computing, which might drive the emergence of breakthroughs in numerous disciplines to tackle previously unsolved challenges. Certain computational problems, such as integer factorization (which forms the basis of RSA encryption), are thought to be solved significantly faster on quantum computers than on classical computers, because quantum computers harness the power of quantum mechanics to deliver massive leaps forward in processing power. The accuracy of the Approximation Ratio after the steps were found to be 89.8

A problem associated with the neural system has been identified as the underlying cause of Alzheimer's disease (AD), a form of dementia. The field of study concerning this matter often relies on clinical data or questionnaires, which can yield conflicting outcomes due to individuals being hesitant to disclose information, possibly due to the nature of the illness or social stigma. Nevertheless, the consequence of early detection of AD cannot be excessive. The existence of recurring or identical characteristics poses a challenge to the efficacy of current computer-assisted procedures and models, which heavily depend on feature extraction. Consequently, achieving optimal results with a single model becomes a formidable task. This study proposes a enhanced way for the classification of Alzheimer's disease utilizing deep learning techniques. The method makes use of texture features and other pertinent information extracted from structural MRI data. Two deep learning models, namely AlexNet and InceptionV3, were employed to construct the predictive model and algorithm. The Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, which categorizes data into five groups, including AD, CN, and MCI, was utilized to evaluate the suggested technique. The sophisticated methodologies employed in this study exhibited superior performance compared to conventional recommendation algorithms in terms of vital assessment parameters utilized to estimate the performance of the supervised classifier. According to the survey results, InceptionV3 emerged as the most successful classifier, achieving the highest accuracy (98.31%) and demonstrating exceptional performance in terms of precision (95.99%), recall (95.81%), and F1 score (95.8%).

In recent years, the exponential growth of online restaurant platforms has revolutionized the dining experience, enabling users to explore various culinary options. Consequently, restaurant rating prediction and personalized recommendation systems have become indispensable tools for users seeking dining suggestions and platform operators aiming to enhance user engagement and satisfaction. The proposed integrated framework for Zomato combines rating prediction with content filtering to optimize user experiences. By integrating machine learning models, the ratings are predicted based on features like location and cuisine type, while content filtering personalizes recommendations. Real-world Zomato data experiments validate our framework’s efficacy in enhancing user satisfaction and engagement, offering tailored suggestions aligned with individual preferences. This work has an accuracy of 0.89 and f1 score of ${0. 8 1}$. Experimental results show significant improvements in rating prediction accuracy and the effectiveness of personalized recommendations, highlighting the potential of our approach to enhance the overall user experience on Zomato.

This research study presents a comprehensive use of machine learning techniques used for bankruptcy prediction, which is crucial for financial institutions to assess creditworthiness. The study highlights methods such as logistic regression, decision trees, and ensemble learning (bagging, boosting), demonstrating the superiority of machine learning over classical statistics for smaller datasets. This study compares various strategies by considering financial statements, market trends, industry benchmarks, and macroeconomic indicators. Integrating diverse data sources helps identify risk factors and signs of financial distress, guiding proactive bankruptcy prevention measures. These include optimizing cash flow, enhancing efficiency, diversifying revenue, and implementing early warning systems. Predicting and preventing bankruptcies support financial stability, job protection, and economic growth. The proposed model employs financial ratios and relevant data to predict bankruptcy likelihood, which is evaluated using assessment and evaluation metrics. The AdaBoost model has a reported accuracy of 86.36% and an F1 score of 85.93%, which are higher than those of all the other cohorts included in the existing work.

With the existing issues related to an employee working for any organization, Attrition is one of the critical issues in the present situation despite the adjustments in the outer condition. It is a steady loss and is supposed to be a slight worker decrease through renunciation, demise, and retirement. When a very much prepared and all-around adjusted worker(employee) leaves the association for any explanation, it makes a vacant space in an association (i.e.) there happens to be a vacuum in the association. It makes it extraordinarily difficult for a Human Resource workforce to fill the hole that has happened. Current Human Resource chiefs are finding a way to lessen the employee wearing down rate. It has been a crucial test for the present managers because they cannot determine the reasons behind any attrition. This paper aims at finding the reasons behind Attrition by applying various machine learning algorithms for making predictions.

Goal: This work focuses on neuroimaging studies, including MRI analysis via machine learning and deep learning, which has led to an increase in computer vision research and the early detection of neural disorders. Methods: An adaptive implementation of transfer learning (TL) with a top-level VGG16 architecture is set up with pretrained weights for large MRI images dataset. Convolutional neural networks (CNN) are a bespoke version of a multi-layer view. Through experimentation on the ADNI dataset, the algorithm was trained and tested in binary and multiclass classification using the MRI scanning of individuals. Results: The machine was trained on the CNN model of binary classification and the CNN model for multiclass classification and was trained using TL (using the VGG16 model) with CNN; 25 epochs of batch size of 32 were considered. To validate the contributions of this study, we demonstrated that the proposed model used for binary classification gave an accuracy of 96.89% and an F1 score of 96.90%, and for multiclass classification, we obtained an accuracy of 99.89% and an F1 score of 94.82%. Conclusion: The suggested approach is highly generic, as it is simple and parameter invariant, and therefore applicable to any MRI dataset.

A stock market has an extreme effect on today’s economy. It has always been a tough job to invest in stock market. In a world where inflation increases at 3% per year, we must invest our money to guarantee our future. The market does not allow us to predict the future of the assets with high accuracy. However, with the successful development in technologies, the opportunities to gain a fortune from the markets are increasing, and it also helps experts predict the stock market. The extensive aim of this paper is to use long short-term memory (LSTM)- based Recurrent Neural Networks (RNNs) to build a machine learning model that can monitor and predict the stock market prices. The performance of this model was determined by accuracy, Root Mean Squared Error (RMSE), and how epochs can modify our model. We evaluated the model using Yahoo Finance data and reported higher accuracy with a prediction of 88.2% in the same cohorts of stocks.

It has been ascertained in recent studies that AD (Alzheimer's disease, a neurodegenerative disorder) and its earlier stages can be detected by neuroimaging biomarkers. To the prevention, previous studies have focused on volumetric asymmetry and brain atrophy. The identification of AD in its early stages has proven to be imperative, as Alzheimer's disease cannot be cured and can only slow its progression. Developing on that idea, this study aims to use the discriminative powers of a Siamese-inspired identical hybrid neural network for the task of classification between multiple stages. The proposed method uses a homemade pipeline for preprocessing and removing other unwanted components from the MRIs that might disturb the model. Registering all the MRI images to MNI space and resampling the slices helped in normalizing the whole dataset. This study uses feature-based methods to work with low-dimensional characteristics rather than high-dimensional voxel data can lessen computing cost and time spent. We used VGG-16 style net with image-net weights for the purpose of automatic feature extraction. T1-weighted MRIs were used for the research, which were accessed from the ADNI datasets ADNI2 and ADNI3. When compared to a normal DNN, our proposed identical hybrid neural network achieved better precision and F1-score.

As of 2020, streaming service giant Netflix has 204 million subscribers. Data experts suggest that Netflix users will spend just 60–90 s browsing and reviewing 10 to 20 titles before finally deciding upon watching content or not. In continuation, a robust recommender system is required to produce user definition-inspired analytics. This paper aims to derive analytical conclusions about cinema data and recommend movies to the user based on their specifications. The problem statement at hand is promptly getting a list of movies to watch next from a dataset of 45,000+ movies. The methods used for analyzing the metadata in cinema are implemented through Python libraries. For building the machine learning model that recommends movies, Natural Language Processing and other relevant statistical/programming like Regular Expression, similarity methods, etc., have been used. The key result of the paper is building a “Content-based recommendation system” based on the user’s definition of similarity.

COVID-19, as the name suggests it is coronavirus disease 2019, comes from severe acute respiratory syndrome (SARS) and middle east respiratory syndrome (MERS) virus family, which is a life-threatening disease. The place of origin of disease has been from country China. It was first spread in Wuhan's city in China and was declared as a pandemic in March 2020 by WHO. To date, no vaccination exists to kill the virus; however, it is being cured and prevented using medicines like hydroxychloroquine, Metformin, dexamethasone, and plasma therapy. In this paper, we have analyzed the state-wise as well as district-wise dataset of country India and classified the states as profoundly affected states and less affected states by performing preprocessing and applying support vector machine model for classification on the state-wise dataset. We also have classified districts into the red, orange, and green zone after analyzing and preprocessing district-wise COVID-19 dataset using Weka Framework.

This paper is focused on creating deep neural networks for the prediction and determination of dementia. It can be used in healthcare, research, and industrial applications. Impairment and dependency are closely related to dementia in older adults. We opted for dementia for our study, as there is no cure present to date for this disorder, and early diagnosis of dementia can help reduce symptoms and delay its progression. In this study, various neural networks were used to classify those affected with dementia from healthy controls by classifying Alzheimer's disease into four stages. For the predictions and classifications, we trained and designed various deep learning models, as the field of deep learning has proven to work well for classification tasks. Convolutional Neural Network, Recurrent Neural Network, and Visual Geometry Group were used to classify healthy controls from the dementia-affected subjects. The results showed that the VGG-16 method had the highest accuracy levels, followed by CNN and RNN. We used Alzheimer's dataset from ADNI on our CNN, RNN, and VGG-16 models. VGG-16 performed with the maximum accuracy of 89.5%, followed by CNN, which had an accuracy of 80.0%, and RNN with an accuracy of 70.2%

Neurodegenerative changes in Parkinson’s disease primarily affect the patient's movement and speech, but they may also cause tremors. This is a central nervous system disease, and neurodegenerative changes mainly affect movement and speech. There are more than 10 million+ people affected worldwide. Its research so far has not yielded any concrete remedies or cures. Researchers believe that through studying past medical data of Parkinson's patients, we can form algorithms that can diagnose the patient with the disease years before symptoms appear. In this research paper, we have analyzed the signs and symptoms facilitating the early diagnosis of Parkinson’s disease by applying classification algorithms on the classical features. The subjects in this study were placed into four classes according to the Unified Parkinson's disease Rating Scale (UPDRS) score after analyzing the various features considered in the above classes. We have attained 87.83 and 98.63

Overview: Parkinson’s disease is a central nervous system neurodegenerative condition that impairs the ability to manage certain bodily activities. Parkinson’s symptoms begin gradually and worsen with time due to dopamine’s supply to the motor system. It is well established that vocal difficulties associated with illnesses may be assessed for early Parkinson’s disease identification. Objective: The primary objective of this project is to develop a categorization system for the typical symptoms of Parkinson’s disease to aid in early identification. Methods: All features were examined and chosen to utilize feature selection algorithms to categorize subjects into four groups based on their UPDRS (Unified Parkinson’s disease Rating Scale) score. SVM, LR, GNB, K-Nearest Neighbor’s (KNN), and Random Forest (RF) were some of the machine learning classifiers that we used to make predictions. Results: SVM had the highest classification accuracy of 89% when combined with PCA for Parkinson’s disease identification.

In this time of a global pandemic, as our society is pushing forward to a more digital culture of working, living, studying, and earning, the Internet's significance is unprecedented, and inevitably so, the use of various social media platforms has seen a boom. This paper focused on hate speech racism and sexism and built a refined classifier that detects racist and sexist comments from the tweets. Investigating accessible data from Internet-based existence may produce interesting findings and bits of information about virtually every object, organization, or individual inside the universe of general feelings. Opinion mining is the statistical action behavior of suppositions, notions, and subjectivity of content from a corpus that integrates natural language processing (NLP) and machine learning (ML) to classify a range of emotions. As such, sentiment analysis remains a widely researched and ever-evolving topic in the field of NLP. XGBoost with word2vec gave us the most promising results, so we refined it further by fine-tuning. Furthermore, we observed that the F1 score was 0.690285. Thus, we were able to reach an accuracy of 69% in our classifier.