Advances in Machine Learning and Big Data Analytics II
ICMLBDA 2023, NIT Arunachal Pradesh, India, May 29-30
- 2025
- Book
- Editors
- Ashokkumar Patel
- Nishtha Kesswani
- Bosubabu Sambana
- Book Series
- Springer Proceedings in Mathematics & Statistics
- Publisher
- Springer Nature Switzerland
About this book
In the dynamic landscape of technology, machine learning and big data analytics have emerged as transformative forces, reshaping industries and empowering innovation. Machine learning, a subset of artificial intelligence, equips systems to learn and adapt from data, revolutionizing decision-making, automation, and predictive capabilities. Meanwhile, Big Data Analytics processes and extracts insights from vast and complex datasets, unveiling hidden patterns and trends. Together, these fields enable us to harness the immense power of data for smarter business strategies, improved healthcare, enhanced user experiences, and countless other applications. This edited volume on machine learning and big data analytics (Proceedings of ICMLBDA 2023, which was held on May 29-30, 2023 by NERIST and NIT Arunachal Pradesh India) introduces an exciting journey into the intersection of machine learning and Big Data Analytics, where data becomes a catalyst for progress and transformation.
Advertisement
Table of Contents
-
Frontmatter
-
Optimized Neural Networks Archetype for Prediction of Socio-economic Class of Women in India
N. M. Jyothi, B. K. Rajya Lakshmi, Pavani Gonnuri, A. PavaniThis chapter explores the optimization of neural networks for predicting the socio-economic class (SEC) of women in India, focusing on a novel architecture that enhances prediction accuracy and speed. The study employs a dataset of 2000 subjects with 30 features, utilizing a Segment, Screen, Integrate, Enhance, and Categorize technique to identify the most impactful features. The research highlights significant improvements in accuracy, achieving up to 99.34%, compared to traditional methods. Key topics include data preparation, feature selection, model architecture, and performance evaluation. The study also compares the proposed model with other machine learning techniques, demonstrating its superior performance in classifying SEC into low, medium, and high categories. The results underscore the model's efficiency in handling large feature sets and its potential for real-time predictions, making it a valuable tool for socio-economic research and policy development.AI Generated
This summary of the content was generated with the help of AI.
AbstractWomen play a vital role in the national economy, and their Socio-Economic Class (SEC) is essential for their empowerment and the nation’s progress. This research aims to build an Artificial Intelligence (AI) model that classifies and compares the SEC of women across different regions of India. Numerous social and economic factors impact women’s SEC, including education, employment, and community support. We employed an Artificial Neural Network (ANN) model, which is enhanced by a new optimization technique involving Segmenting, Screening, Integrating, Enhancing, and Categorizing. This innovative method makes more precise feature selection possible and improves computational efficiency. The research outcomes reveal that this innovative technique significantly boosts the ANN’s performance, achieving an impressive classification accuracy of 99.34%. Compared to previous work, this model performs exceptionally well. -
Improvement of UPFC Performance in Power Systems Using Artificial Intelligence
Mamatha Deenakonda, Padma Jyothi Uppalapati, Sridevi Bonthu, Phanikumar ChittalaThis chapter delves into the enhancement of UPFC performance in power systems through the application of Artificial Intelligence, specifically Artificial Neural Networks (ANNs). The study identifies key challenges in power transmission, such as higher Total Harmonic Distortion (THD), voltage fluctuations, and stability issues, and proposes an innovative solution using ANN-based controllers. The integration of UPFC with ANN controllers aims to achieve near-unity power factor, reduce THD, and improve power quality. Simulation results demonstrate significant reductions in THD levels, with values as low as 0.25% for linear loads and 0.3% for nonlinear loads, highlighting the effectiveness of the proposed system. The chapter also explores the role of UPFC as an active power filter, enhancing power quality and stability in transmission lines. The detailed analysis and simulation results provide a comprehensive overview of the benefits and practical applications of ANN-based UPFC systems in modern power networks.AI Generated
This summary of the content was generated with the help of AI.
AbstractTo address various challenges in power generation and distribution such as power quality deterioration, total harmonic distortion (THD), stability issues, and transient response, FACTS (Flexible Alternating Current Transmission Systems) devices are commonly employed. The Unified Power Flow Controller (UPFC) plays a significant role among these devices. Artificial Intelligence techniques, particularly Artificial Neural Networks (ANNs), are utilized to enhance their performance. In this paper, we propose a novel approach combining an ANN controller with UPFC. This combination aims to significantly improve upon previous methodologies by reducing THD values and enhancing power system stability. In conjunction with a shunt filter and transformer, the ANN controller effectively manages power transient conditions, thereby demonstrating superior performance compared to conventional methods. -
Classification of Parkinson’s Disease Using Machine Learning Techniques
Satish Dekka, K. Narasimha Raju, D. Manendra Sai, M. Pallavi, Bosubabu SambanaThis chapter delves into the application of machine learning techniques for the early detection and classification of Parkinson's disease, focusing on both motor and non-motor symptoms. The study evaluates the effectiveness of three machine learning algorithms—decision trees, support vector machines (SVM), and K-nearest neighbors (KNN)—in predicting the progression of Parkinson's disease. The research utilizes a dataset from the UCI repository, with variables selected based on symptoms and diagnosis. The performance of the algorithms is assessed using various metrics, including accuracy, area under the curve (AUC), and receiver operating characteristic (ROC) curves. The results indicate that the KNN algorithm achieves the highest accuracy of 91.9%, outperforming decision trees and SVM. The chapter also discusses the potential of these machine learning techniques in improving the early detection and management of Parkinson's disease, highlighting the importance of proactive treatment to enhance disease outcomes. Additionally, the study explores the impact of Parkinson's disease on various body parts and the role of machine learning in analyzing different reports and identifying diseases. The comparative analysis of the algorithms provides insights into their suitability for predicting Parkinson's disease, making this chapter a valuable resource for professionals seeking to leverage machine learning in neurological research.AI Generated
This summary of the content was generated with the help of AI.
AbstractEmotional speech and adverse social effects, such as stigma, dehumanization, and loneliness, are the main issues with Parkinson’s disease. This has a significant impact on the way of life of those who have Parkinson’s disease. The slow death or destruction of brain neurons is one of the known causes of Parkinson’s disease. As a result, dopamine is produced in the brain as a chemical messenger. Dopamine deficiency results in typical brain activity and a variety of modifications in how the body moves. Parkinson’s disease (PD) is mostly recognized by its signs and symptoms, such as a little tremor. Body movement is slowed down by tremors. Rigid muscles in this body weaken and lose automatic movements like blinking and smiling, which are additional symptoms. Previous researchers are still working on this problem, but they are having trouble using the right techniques to solve it. The suggested work utilized the MATLAB environment to develop three machine learning algorithms. The outcome indicates that the KNN algorithm has the highest accuracy in detecting Parkinson’s illness. -
A Survey on Skin Cancer Detection Using Artificial Intelligence
N. P. Patnaik M, Johan Jaidhan BeeraThis survey delves into the application of artificial intelligence in skin cancer detection, focusing on the use of neural networks for image classification. It examines the stages of skin disease recognition, including preprocessing, image segmentation, feature extraction, and classification, with a particular emphasis on the performance of artificial neural networks (ANNs), convolutional neural networks (CNNs), k-nearest neighbors (KNNs), and radial basis function networks (RBFNs). The survey highlights the superior results achieved by CNNs in image classification tasks due to their close connection with computer vision. It also explores the emerging concept of auto-association in deep learning, which aims to identify features and patterns in image datasets to enhance the accuracy of diagnostic systems. The text discusses the current limitations of AI-based diagnostic tools, such as their focus on individual lesion images rather than full-body assessments, and the challenges posed by the low contrast and heterogeneous backgrounds in dermoscopic images. It also reviews various studies and methodologies, including the use of statistical models for wrist bone segmentation in CT images and the development of mobile device-based systems for dermoscopic image classification. The survey concludes by emphasizing the importance of early detection in improving survival rates for skin cancer and the need for further research to address the complexities of automated skin lesion analysis.AI Generated
This summary of the content was generated with the help of AI.
AbstractSkin cancer ranks among the worst types of disease. For skin cancer to occur, there must be a mutation in the skin’s genetic material or a flaw in the DNA of skin cells. The importance of early skin cancer detection cannot be overstated since it can spread to other parts of the body and thus be less treatable later. It is important to recognize early warning signs of skin cancer because it is prevalent, has a high mortality rate, and is expensive to treat. Considering the gravity of these concerns, scientists have devised a number of methods for detecting skin cancer in its earliest stages. Skin cancer may be detected and classified as either benign or malignant based on several lesion factors, including but not limited to symmetry, color, size, form, etc. This study provides a comprehensive overview of AI methods for spotting skin cancer in its earliest stages. High-quality research publications on skin cancer diagnostics were reviewed. Tools, graphs, tables, methods, and frameworks portray research findings. -
A Review of Ranking Approach to Rank Interval-Valued Trapezoidal Intuitionistic Fuzzy Sets
S. N. Murty Kodukulla, V. SireeshaThis chapter delves into the application of the Centre of Gravity (COG) method in ranking Interval-Valued Trapezoidal Intuitionistic Fuzzy Sets (IVTrIFS), a crucial aspect of fuzzy logic and decision-making processes. The author introduces a new ranking approach that leverages the COG of the hesitancy region, a mathematical expression quantifying uncertainty in decision-making. The hesitancy region is divided into three parts, and the COG is calculated through these regions to define a ranking function based on the Euclidean distance. The chapter provides a detailed geometric representation and mathematical formulation of the hesitancy region, including the areas and COGs of the constituent triangles and rectangles. A numerical example compares the proposed method with existing ranking methods, demonstrating its consistency and improved performance in specific cases. The conclusion underscores the significance of the proposed ranking method as an alternative approach for ranking IVTrIFSs, highlighting its alignment with certain existing methods and its advantages over others.AI Generated
This summary of the content was generated with the help of AI.
AbstractAn Interval-valued trapezoidal intuitionistic fuzzy set (IVTrIFS) is a powerful tool in modeling uncertainty. It is a special type of Intuitionistic Fuzzy Set (IFS) and interval-valued intuitionistic fuzzy set (IVIFS), which has a consecutive domain of real numbers. An IVTrIFSs is distinguished by using three characteristic functions namely membership degree, non-membership degree, and hesitancy degree. Ranking is a challenging task, and every ranking method has its own significance and applicability. Since their inception in 1965, several researchers have proposed various ranking methods by using concepts such as Score and Accuracy, Centroids, Centre of Gravity, Distance/Similarity measures, Value and Ambiguity, etc. yet there is no specific ranking approach that is appropriate for all applications or provides adequate results in all situations. The Centre of Gravity (COG) method is one of the most popular defuzzification techniques of fuzzy mathematics. By utilizing the notion of the COG, one can ascertain the central tendency or representative value of a given fuzzy set. As it is derived by using area and centroids of any geometric representation, this method is particularly useful in fuzzy logic systems, fuzzy control, decision-making processes, and fuzzy data analysis. In this paper, we derive a ranking method for IVTrIFS using the Centre of Gravity. -
An Empirical Evaluation of ResNet-SE-16 for Accurate Classification of Lung Cancer Using Histopathological Images
P. Vishal, D. Kishore Kalyan Kumar, K. Venu Kiran Raju, K. Jayalaxmi Manojna, Mohan MahantyThis chapter explores the development and evaluation of an enhanced ResNet-SE-16 model for classifying lung cancer subtypes using histopathological images. The study focuses on improving computational efficiency while maintaining high accuracy. Key topics include the model's architecture, data acquisition and augmentation, training and testing procedures, and performance metrics. The model achieves 97% accuracy, outperforming other models in terms of F1 score, precision, recall, and accuracy. The integration of Squeeze-and-Excitation Networks (SENets) with a 16-layer ResNet allows for efficient feature extraction and classification. The model's performance is validated using a confusion matrix, demonstrating its effectiveness in distinguishing between adenocarcinoma, squamous cell carcinoma, and benign tissue. This research highlights the potential of enhanced ResNet models in medical imaging, offering a balance between accuracy and computational efficiency.AI Generated
This summary of the content was generated with the help of AI.
AbstractLung cancer is a leading cause of cancer-related deaths worldwide, accounting for approximately 25% of all cancer-related deaths. It is much more deadly than colon, breast, and prostate cancers combined. Early detection and treatment of cancer is crucial for a patient’s recovery. Radiologists use histopathological images to diagnose potentially infected lung regions, but this process can be time-consuming. Deep learning, a type of machine learning, can hasten this process by simulating how the human brain functions. Convolutional Neural Networks (CNNs) can quickly and accurately classify and identify the various kinds of lung cancer, improving patient treatment and survival chances. While many CNN models have been developed by researchers, they often require significant computational power and time. In this project, we developed an improved CNN model that classifies lung cancer types with high accuracy while requiring less computational power than other models. Using metrics like accuracy, precision, recall, and F1 score, our model obtained an accuracy of 98% during training and 97% during validation. -
Design of EV Battery System Using Grid-Interface Solar PV Power with Novel Adaptive Digital Control Algorithm
C. Sudhakar, K. Keerthi Reddy, D. Gopal Krishna, P. Pavan, N. Sreelatha, K. RohiniThis chapter delves into the integration of solar photovoltaic (PV) power with electric vehicle battery (EVB) systems, focusing on grid stability and efficiency. The text explores the use of a novel adaptive digital control algorithm to optimize power quality and system performance. Key topics include the implementation of a single-stage converter design, which reduces costs and increases efficiency, and the application of a perturb and observe (P&O) method for maximum power point tracking (MPPT). The chapter also discusses the challenges of integrating renewable energy sources with the grid, such as harmonics injection and voltage fluctuations, and presents solutions to mitigate these issues. Furthermore, the text examines the role of battery storage in managing the intermittent nature of solar PV power and enhancing grid stability. The conclusion highlights the successful implementation of the adaptive recursive digital filter control technique, demonstrating improved power quality and dynamic performance under varying conditions.AI Generated
This summary of the content was generated with the help of AI.
AbstractThis inquiry aims to show a solar photovoltaic (PV) system that is linked to the grid and includes characteristics that allow for adjustments to be made to the power quality. Three major phases may be identified within the system. This system is employed to compensate for a range of power quality (PQ) issues, including harmonics, redundant reactive power, and load unbalancing, in addition to transferring power that is produced by a photovoltaic (PV) array to feed linear and nonlinear loads. This system also feeds linear and nonlinear loads. These are only some of the problems that our technology could be able to solve in the future. In addition to this, it can provide linear power to loads when this kind of power is required. A three-phase voltage source converter (VSC) is used so that the direct current (DC) electricity that is produced by the PV array may be converted into alternating current (AC). Alternating current, or AC, is the kind that is most often used. It is essential for the solar PV system that is connected to the grid to have an effective control strategy in order to simplify the transmission of active energy and to reduce the potential that power quality issues will occur. This research aims to illustrate how an adaptive generalized maximum Versoria criteria (AGMVC) controller may be used for a variable speed drive (VSC), which is a component of a solar photovoltaic energy conversion system. To guarantee effective utilization of the solar photovoltaic array, the maximum power point tracking (MPPT) method, which is founded on the perturb and observe algorithm, is used. The grid-integrated PV system’s experimental environment is first fabricated in the laboratory using an IGBT-based VSC and DSP (dSPACE DS-1202), and then the system itself is put together. Experiments are carried out to determine how effective the AGMVC control approach is. These experiments make use of a prototype that was developed inside the laboratory. This control method is evaluated in comparison with a variety of different conventional controllers, such as synchronous reference frame theory (SRFT) and instantaneous reactive power theory (IRPT), in addition to recently developed weight-based controllers, such as least mean square (LMS), least mean mixed norm (LMMN), and normalized kernel least mean fourth-neural network (NKLMFNN). To make a comparison between AGMVC and the control methods that have been discussed in the past, a number of criteria, such as fundamental weight convergence, steady state error, computational complexity, the requirement for phase lock loop (PLL), and the potential for providing harmonic compensation, are taken into consideration. The purpose of this comparison is to establish whether the AGMVC is more effective than the control approaches that were discussed before. In accordance with the IEEE-519 standard, the functionality of the system is evaluated and ranked in accordance with its performance during the testing. Harmonics, the maximum Versoria criteria, power factor correction (PFC), and solar photovoltaic are just some of the index terms that may be discovered in this section. -
A Decentralized and Intelligent Approach for Suspicious Event Detection in Surveillance Range
K. U. V. Padma, E. NeelimaThis chapter delves into the integration of blockchain technology and the Inter-Planetary File System (IPFS) to create a decentralized and intelligent approach for suspicious event detection in surveillance systems. The paper explores the use of smart contracts for access control and the implementation of a certificate authority to monitor user behavior, enhancing data security and mitigating risks of unauthorized access or data manipulation. The system framework involves two primary entities: the Data Owner Group (DOG) and the Data User Group (DUG), with the smart contract deployed on the blockchain to facilitate secure data sharing. The paper also discusses the advantages of using IPFS for decentralized storage, including reduced bandwidth requirements and the ability to host websites. Furthermore, the chapter provides a comprehensive literature survey, highlighting the applications of blockchain technology in various industries and the advantages of decentralized storage systems. The conclusion emphasizes the practicality of storing encrypted files on IPFS due to blockchain's limitations in handling large data sets, and the use of smart contracts for verifiability and ensuring service fees are only deducted upon achieving the desired result.AI Generated
This summary of the content was generated with the help of AI.
AbstractAs the world becomes progressively more concerned about security, a gigantic amount of observational data is being generated by security systems. This data, produced on a daily basis, presents challenges for companies looking to store and manage it efficiently. The scattered nature of the data presents a critical challenge to data storage and analysis. One major concern is the security of data exchanged over devices, as any adversary can capture or alter the data, deceiving the reconnaissance system. Users can exchange files and data over the globe due to the scattered nature of the arrangement. Nevertheless, large files consuming a large amount of transmission bandwidth to upload and download over the web can benefit from the use of IPFS, which has rapidly gained notoriety due to its capacity to run the best of multiple protocols, including FTP and HTTP.Despite IPFS’s numerous focal points, there are security and access control issues, including a lack of traceability in how files are accessed. To address these issues, this article proposes a novel procedure that enhances IPFS with blockchain technology to give a straightforward audit trail. By leveraging blockchain as an asset, data reliability and source security can be improved, offering a clear way to trace all activity related to a specific file. This approach can enhance the security and integrity of the data, increasing confidence in the surveillance system’s ability to work effectively and securely. -
Image Selection for Graphical Password Authentication
P. Anjaneyulu, D. Priyanka, T. Chalapathi, B. Samanvi, B. MounikaThis chapter delves into the world of graphical password authentication, presenting it as a compelling alternative to traditional alphanumeric passwords. The text explores the limitations of conventional password systems, including their vulnerability to brute force attacks and the challenges users face in remembering complex passwords. It introduces graphical passwords as a solution that leverages the human ability to recognize and remember images, offering enhanced security and usability. The chapter discusses various graphical password techniques, such as recognition-based and recall-based approaches, and provides a comprehensive survey of existing methods. It also outlines the implementation of a graphical password system using Django, including user registration, image selection, and login processes. The text highlights the advantages of graphical passwords, such as increased security, ease of use, and reduced risk of password reuse. Additionally, it addresses challenges like user memorability, image set selection, and system performance. The chapter concludes with a practical example of a graphical password authentication system for web-based applications, demonstrating the potential of this innovative approach in enhancing user authentication security.AI Generated
This summary of the content was generated with the help of AI.
AbstractOne of the most crucial elements of information security is user authentication. User authentication is the primary strategy for ensuring the utmost factors for computer security. It gives the base for access control and user liability. Over time, among numerous types of user authentication systems, alphanumeric passwords have remained the most common. A graphical password is a type of authentication system that relies on the user’s selection of images presented in a graphical user interface (GUI), in a specific order. This approach, known as graphical user authentication (GUA), differs from the traditional system of using alphanumeric usernames and passwords, which is presently the most common way of authenticating computer users. This system has been shown to have significant disadvantages. For example, users tend to choose passwords that can be fluently guessed. On the other hand, if a password is delicate to guess, then it is frequently difficult to remember. To overcome this problem of low security, authentication styles are developed by experimenters, using images as a password. This explorative paper is a detailed examination of current graphical passwords, proposing a new proposition. Graphical passwords have been proposed as a volition to textbook-grounded schemes, based on the idea that humans tend to remember pictures more fluently than text. Also, pictures are considered to be more user-friendly. -
Soft Computing for Visual Recognition Through Audio for the Visually Impaired
S. R. K. L. Amulya, Vishakha Singh, Tummalapalli Sandeep, Surya Sasidhar, Rita RoyThis chapter delves into the application of soft computing techniques for visual recognition through audio to assist the visually impaired. The study evaluates various machine learning models, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), to enhance mobility and independence for individuals with visual impairments. The research focuses on the use of CNNs for feature extraction and RNNs for generating descriptive captions, which are then converted to audio output. The study compares different CNN models, such as VGG16, ResNet, InceptionNet, XceptionNet, and MobileNet, in combination with GRU, a variant of RNN, to determine the most effective approach. The results indicate that the XceptionNet-GRU model achieved the highest BLEU score of 0.86, demonstrating its superior performance in generating accurate and descriptive captions. The chapter also discusses the practical applications of these models in real-world scenarios, highlighting their potential to improve the quality of life for visually impaired individuals. The study concludes by suggesting future investigations to enhance the system's capabilities, such as using larger datasets and real-time image descriptions.AI Generated
This summary of the content was generated with the help of AI.
AbstractVisually impaired people are frequently unaware of outside obstructions and require assistance to avoid dangerous collisions. The sense of sight is the most fundamental aspect of human perception and plays. We propose an object recognition algorithm and assistance system that is very useful for their safety and quality of life. This project aims to use an available mobile device as a talking guide dog to aid those with vision impairments in outdoor navigation and inform them about obstacles. The proposed technology will lower the risks of obstacle contact by allowing users to move outside without stumbling, potentially informing the user what the objects are. The approach proposed uses algorithms of deep learning. CNN recognizes salient functions and captions of photos and converts written text to speech by detecting features through the broadcasted picture alongside its relevant caption, while the Gated Recurrent Unit (GRU) network acts as a tool that captions and describes the text detected from photographs. The predicted caption is transformed into an audio message. The proposed Convolution Neural Networks [CNNs] Gated Recurrent Units [GRUs] model has investigated the usage of numerous network architectures: Inception Net, Mobile Net, Xception Net, ResNet, and VGG16. -
Diet Meal Plan Chatbot
Saubhagya Jung Karki, Sarthak Pokhrel, Saugat Chand Thakuri, Anupam Pandey Chhetry, Yashpal SinghThis chapter delves into the transformative potential of AI-powered chatbots in personalized diet planning and weight management. It covers the integration of Natural Language Processing (NLP) and machine learning algorithms to create tailored meal plans based on user-specific data such as BMI, BMR, and dietary preferences. The chapter also explores the technical architecture of the chatbot, including the use of feedforward neural networks, encoder-decoder models, and stochastic gradient descent for intent classification and response generation. Additionally, it discusses the importance of user feedback and the chatbot's ability to adapt and improve over time. The chapter concludes with a comparison of the proposed chatbot with existing solutions, highlighting its unique features and advantages. By reading this chapter, professionals will gain insights into the latest advancements in AI technology and its practical applications in health and fitness.AI Generated
This summary of the content was generated with the help of AI.
AbstractA healthy life is an important issue that has been affecting a large part of the human population as the obesity rate started to rise persistently while people have been rebuking it instead of finding a solution. We need to find out the method before having stinging defeat. Currently, AI-based software for personalized nutrition is getting hyped up for users to gain a healthy lifestyle. Chatbots empowered by Artificial Intelligence (AI) can be solutions to engage with humans in natural conversation and build relationships in the era of the Internet. Here in this paper, we present the knowledge of a Chatbot called “Diet-meal Chatbot” built by us to pitch it on the perfect line for the betterment of human beings. A diet meal chatbot is a conversational agent designed to have a tailored diet management plan. It uses NLP, a machine learning algorithm to understand user preferences and diet requirements, and suggests a balanced meal plan. It will be able to give them highly accurate physical exercise, healthy food style, and weight management style, for both males and females under whichever age. Chatbot will be able to suggest food for users based on their food intake, water intake, activity level, weight of users, age, and height as well. -
Convolution Neural Networks
K. Bhagyalaxmi, B. DwarakanathThis chapter explores the application of convolutional neural networks (CNNs) for the automated detection and classification of brain tumors from MRI images. The study focuses on the challenges of manual tumor detection, the role of deep learning in medical imaging, and the implementation of a VGG-19-based transfer learning approach. Key topics include the variability in tumor regions, a literature review of existing methods, and the methodology behind image recognition and CNN architecture. The implementation details cover dataset sources, data preprocessing, and the training procedure using the VGG-19 model. Results demonstrate an impressive accuracy of 84.31% on the validation set and 98.75% on the training set, with classification metrics and confusion matrices provided for different tumor types. The conclusion highlights the superior performance of the VGG-19 architecture and suggests future directions for improving accuracy and expanding the model's applications.AI Generated
This summary of the content was generated with the help of AI.
AbstractBrain tumors are extremely dangerous and can lead to a very short life expectancy at their highest grade. Identifying brain tumors early is crucial for improving survival rates, but it poses a significant challenge for medical professionals. Noise and other environmental disturbances are more likely to appear in MRI. As a result, it is challenging for doctors to diagnose the tumor and determine its causes. Therefore, a system was proposed to detect brain tumors from images. In this procedure, the image is first converted into a grayscale then filters are applied to remove environmental interference and other noise from the image. The user has to select the image. The system processes images through various image-processing steps. However, in the early stages of a brain tumor, the edges in the images are not well-defined. To address this, a deep learning model incorporating CNN and FCNN has been developed to enhance the classification of tumor types and the associated preprocessing stages. This pre-trained network is capable of categorizing images into 1000 object categories. The fully connected layers are replaced with dense connections and a softmax activation function, enabling the classification of brain tumor images into four classes. -
Segmenting MRI Images Using Federated Learning for Brain Tumor Detection
T. L. Akshaya, K. Swetha, S. Pravalika, U. ChandrasekharThis chapter delves into the application of federated learning for segmenting MRI images to detect brain tumors, a critical task given the rising prevalence of brain tumors worldwide. The study focuses on two advanced deep learning models, U-Net and DeepLabV3, and evaluates their performance in a federated learning setup. The BraTS2020 dataset, known for its high-quality MRI scans and detailed annotations, serves as the foundation for this research. The paper explores various federated learning strategies, including different initial weight loading methods and weight averaging techniques, to optimize model performance while preserving data privacy. The results indicate that federated learning can achieve high accuracy in MRI image segmentation, with U-Net showing promising results in predicting segmented mask images. The study also discusses the legal and ethical implications of handling medical imaging data, highlighting the importance of adhering to data usage policies and regulations. Additionally, the chapter suggests future research directions, such as leveraging metadata for tumor type prediction and implementing advanced security techniques during data sharing. This comprehensive analysis provides valuable insights into the potential of federated learning in medical imaging, offering a balanced approach to improving diagnostic accuracy while safeguarding patient privacy.AI Generated
This summary of the content was generated with the help of AI.
AbstractAI systems must be able to make use of large, diverse, and global information in order to build robust and efficient systems in the medical imaging field. To develop a global model, all these facts must be collected in one place, but this raises questions about privacy and ownership. In this study, we evaluated multiple federated learning methods for segmenting brain tumors. Federated learning utilizes all accessible data without storing or disclosing collaborators’ personal information on a central server. By appropriately integrating these model updates, a high degree of accuracy could be attained, but doing so increases the possibility that the shared model might accidentally leak the local training data. We have selected the best model from U-Net and DeepLabV3 to obtain the best efficiency. -
Comparative Analysis of Fuzzy Regular Graph Properties
Nagadurga Sathavalli, Tejaswini PradhanThis chapter delves into the fascinating world of fuzzy regular graphs, presenting a comparative analysis of their properties. It begins by defining fuzzy graphs and exploring the concepts of regularity and total-degree in these graphs. The text provides a necessary and sufficient condition for the equality of regular and completely regular fuzzy graphs, supported by various examples. It also investigates the properties of regular fuzzy graphs with a cycle as the underlying crisp graph. Additionally, the chapter discusses the vertex degree in different fuzzy graph products and operations, such as union, join, Cartesian product, composition, and conjunction. The analysis is further enriched by exploring fuzzy rule extraction and fuzzy composition. The chapter concludes with a discussion on the union of fuzzy sets and the properties of complete regular fuzzy graphs, providing a comprehensive overview of the topic.AI Generated
This summary of the content was generated with the help of AI.
AbstractWe are going to discuss fuzzy regular graphs and their properties, such as fuzzy rules, cartesian products, and the If-Then rule. A very helpful tool for describing crucial aspects of a physical system with finite components is the fuzzy regular graph. In this work, we explore the fascinating results of the IF-THEN rule using a new parameter for both regular and completely regular fuzzy graphs via numerous examples, regular fuzzy graphs and completely regular fuzzy graphs are compared. -
Effective Cloud Data Management by Using AES Encryption and Decryption
Dogga Aswani, D. Jaya Kumari, Alluri Neethika, G. Sujatha, Praveen Kumar Karri, Sowmya Sree KarriThis chapter delves into the critical aspects of cloud data security, focusing on the implementation of the Advanced Encryption Standard (AES) algorithm. It highlights the vulnerabilities of the older Data Encryption Standard (DES) and demonstrates why AES is a superior choice for protecting sensitive data in cloud environments. The text provides a comprehensive overview of the AES encryption and decryption processes, including key expansion, substitution, permutation, and round key addition. It also discusses the experimental setup using Java and a hybrid cloud server, showcasing the performance benefits of AES over DES. The chapter concludes with a user-friendly interface for file encryption and decryption, emphasizing the practical applications of AES in real-world scenarios. Readers will gain insights into the step-by-step procedures of AES implementation, the advantages of using larger key sizes, and the importance of continuous security monitoring in cloud environments.AI Generated
This summary of the content was generated with the help of AI.
AbstractData security is a crucial issue in today’s digital world of communication. As we all know, hackers are everywhere these days, looking for valuable information that they can exploit for various purposes. When it comes to government data from any country, the risk doubles. As a result, a system or terminology must be in place to keep data secure at all times throughout transmission. Data protection can be achieved by converting the original data to some other unusable data, so that if that data is obtained, it must likewise remain in unusable bits. Previously, the DES block cipher algorithm was employed for security, but it had some limitations. DES has a key length of 56 bits and can be compromised by brute-force attacks. It is also subject to attacks that use linear cryptanalysis. So, we propose the AES Stream cipher technique for increased security, with key lengths of 128, 192, and 256 bits that will not be compromised by brute force assaults. It is more efficient than the DES cipher. In this proposed work, we aim to apply the AES method for encrypting and decrypting the data that is placed on the cloud server so that we may store the data effectively and efficiently for cloud users. -
Prediction of Angina Pectoris
D. Dakshayani Himabindu, Raswitha Bandi, Keesara Sravanthi, V. Sai Vandana, G. Uday BhaskarThis chapter delves into the prediction of angina pectoris, a critical symptom of coronary heart disease, using machine learning techniques. The study explores the potential of replacing traditional diagnostic methods, such as ECG and MRI tests, with a more efficient, digital solution. The research investigates various machine learning algorithms, including logistic regression, support vector machines, random forest, and XGBoost, to identify the most effective method for predicting angina. The study achieves an impressive accuracy rate of 94% using the random forest algorithm. The chapter also discusses the implementation of a chatbot-like platform to provide a user-friendly interface for patients. The research highlights the importance of early detection and the potential of machine learning in improving healthcare outcomes. The study concludes with a discussion on the future scope of the project, including the integration of additional diseases and personalized nutrition recommendations.AI Generated
This summary of the content was generated with the help of AI.
AbstractAngina pectoris is the clinical term for chest discomfort associated with cardiovascular disease. It happens when the cardiac muscle is not getting enough oxygenated blood. Ischemia is a disorder in which some or all of the heart’s arteries are constricted or blocked. Many persons who are originally suspected of having angina have normal coronary angiograms, indicating that they do not, in fact, have angina. A feasibility study was conducted to determine the practicality of a preliminary screening test. Information on a variety of possible risk factors was collected for a large number of patients suspected of having angina, and then their angina status was documented. The major goal of this research was to see which health characteristics, if any, are linked to angina and whether a subset of them might be utilized to predict the dependent variable angina status. More specifically, being able to estimate the risk/probability that a person with a specific mix of these health indicators has angina would be beneficial. Furthermore, estimating the individual effects of relevant factors would be interesting. -
A Novel Web Attack Recognition System for IOT via Ensemble Classification
Preethi Bitra, Kandukuri Rekha Sai KumarThis chapter delves into the critical need for advanced security measures in IoT devices, focusing on the development of a novel web attack recognition system. The study explores the limitations of existing systems and proposes an ensemble classification model to enhance accuracy and efficiency in detecting fraudulent activities. Key topics include a literature survey of relevant research, an overview of the proposed system, and a detailed analysis of the dataset used for testing. The chapter also compares the performance of various machine learning models, such as SVM, Decision Trees, and Random Forest, with the Random Forest model emerging as the most effective. The experimental results demonstrate the system's high accuracy and efficiency, making it a promising solution for securing IoT devices against web attacks. By integrating multiple machine learning techniques, the proposed system offers a robust framework for continuous user identification and fraud detection, paving the way for more secure IoT environments.AI Generated
This summary of the content was generated with the help of AI.
AbstractMachine learning is increasingly utilized in various fields such as biomedicine, information security, firewall protection, medical science, the Internet of Things (IoT), and mobile application security. Consequently, the demand for machine learning applications is growing. In this article, we apply machine learning to detect web or Internet attacks on IoT mobile devices. With the rapid growth in smartphone usage, protecting personal information and privacy has become a significant challenge. Impersonation is a major consequence of data leaks. Current preventive methods, like passcodes and fingerprinting, cannot continuously monitor usage to verify user authorization. Typically, once a user is authorized, they gain complete control of the device, and no further protection can prevent access to device data. To address this, we propose an ensemble machine learning model for detecting impersonation, based on a multi-view bagging strategy that collects sequential tapping information from the smartphone keyboard. This model continually authenticates the user while typing using sequential-tapping biometrics. By conducting multiple tests on our model, we evaluated it using the CLaMP (Classification of Malware with Portable headers) dataset from Kaggle. Our empirical and theoretical results demonstrate that our model outperforms other approaches, achieving a 6.42% equal error rate, 93.14% accuracy, and 95.41% H-mean using only the accelerometer and five keyboard taps.
- Title
- Advances in Machine Learning and Big Data Analytics II
- Editors
-
Ashokkumar Patel
Nishtha Kesswani
Bosubabu Sambana
- Copyright Year
- 2025
- Publisher
- Springer Nature Switzerland
- Electronic ISBN
- 978-3-031-51342-8
- Print ISBN
- 978-3-031-51341-1
- DOI
- https://doi.org/10.1007/978-3-031-51342-8
PDF files of this book have been created in accordance with the PDF/UA-1 standard to enhance accessibility, including screen reader support, described non-text content (images, graphs), bookmarks for easy navigation, keyboard-friendly links and forms and searchable, selectable text. We recognize the importance of accessibility, and we welcome queries about accessibility for any of our products. If you have a question or an access need, please get in touch with us at accessibilitysupport@springernature.com.
