Manju Khari

This book provides an in-depth exploration of the integration of cloud and edge computing in artificial intelligence, addressing critical topics such as hybrid

The integration of edge computing into smart healthcare systems offers a transformative approach to addressing the growing demands for low-latency, secure, and reliable medical data processing. With the rapid expansion of Internet of Medical Things (IoMT) devices and AI-driven diagnostic tools, traditional cloud computing often fails to deliver the required real-time responsiveness and data privacy. This study investigates the potential of edge computing as an efficient alternative by developing a prototype healthcare setup where IoMT devices operate through localized edge servers. The performance of this edge-based framework was compared to a conventional cloud model using key metrics such as latency, energy efficiency, throughput, data security, and overall quality of service. Data Envelopment Analysis (DEA) was employed to evaluate the relative efficiency of both systems, considering each configuration as a decision-making unit. DEA enabled a multi-dimensional assessment linking technical inputs, such as CPU utilization and latency, to outcomes like response speed and security. The findings reveal that edge-enabled architectures significantly reduce response time, improve data confidentiality, and enhance system reliability in critical healthcare operations. Thus, edge computing, when integrated with AI and IoMT, provides a scalable, privacy-aware, and real-time framework for next-generation smart healthcare applications.

The evolution in the Edge AI data processing capabilities directly over edge devices by using intelligent computation on the device itself paves the way for finding new measures to incorporate security and privacy considerations. Edge AI provides numerous advantages, such as reduced latency, increased bandwidth, and a straightforward user interface. Therefore, new challenges related to security and privacy become highly crucial. As in Edge AI systems, all the sensitive data is processed locally, so they are susceptible to physical tampering, adversarial attacks, malware exploitation, and eavesdropping-kind of vulnerability. The inherent distributed nature of edge devices presents significant challenges when an architecture designer tries to form a cohesive security framework for edge devices forming the critical application framework. This complexity of the local modelling at edge devices itself necessitates the need for developing robust strategies where the prime focus was laid on fundamental security measures themselves. Some of the pivot measures are using strong multi-factor authentication (MFA) processes to verify the identity of users and devices along with using strong data encryption algorithms to safeguard sensitive information so that it can be protected from unauthorised access by threat actors. In addition to the above, we have to identify and use secure communication protocols for data transmission between devices so that it remains confidential and tamper-proof, and hence provide stealth to the overall security posture in edge computing environments. At the same time, the privacy concerns have increased because edge devices tend to collect and process personal and sensitive information like patient health records in smart health monitoring systems and location data in smart transportation and smart city applications. This raises the need for compliance with regulations such as General Data Protection Regulation (GDPR), which is a European Union law governing the processing and sharing of personal data, and Health Insurance Portability and Accountability Act (HIPAA), a set of rules designed to protect the privacy and security of patients’ health information, and DPDP Act 2023, which is an Indian law applicable to the processing of digital personal data within India. These rules govern all the data collected online or offline, which further goes under digitisation at the edge devices themselves by local AI models. These regulations need to be enforced on healthcare providers, smart city developers, smart transportation entities, etc. Furthermore, due to the unique challenge posed by machine learning attacks, capabilities checks and balances need to be incorporated by continuous monitoring and regular enhancements in local model capabilities. There are many emerging technologies available like hardware-based approaches, federated learning, and trusted execution environments which can enhance privacy and reduce exposure to breaches by providing decentralised model training without sharing raw data along with a multi-layered approach to protect against diverse threats.

Financial fraud has become a prevalent problem and with the word ‘fraud’, credit card fraud is usually perceived by everyone. Credit cards have long been in use; the large amount of money that has been stolen from accounts and subsequent losses to banks and businesses has become a pressing concern. The escalating threat of credit card transaction fraud poses significant implications for the financial sector. Data plays a pivotal role in detecting both online and offline credit card fraud, yet it presents challenges due to the evolving nature of fraudulent techniques and the inherent asymmetry of credit card fraud datasets. There are numerous algorithms developed to identify whether a transaction is fraudulent or not. In this study, different algorithms have been tested on the data of European cardholders containing 93,181 entries of transactions. In this study, the five algorithms have been implemented in python language and the effectiveness of each algorithm tested based on accuracy, precision, recall rate and F1-score. The research identified that the most effective classification algorithm for credit card fraud detection is XGBoost with accuracy of 96.96% and F1 score of 91.95%. By providing insights into the performance of different algorithms in handling imbalanced data and detecting fraudulent transactions, this study contributes to the ongoing efforts to combat credit card fraud in the financial sector.

Botnets represent a significant cyber security risk, as such networks of compromised devices under attacker control can wreak havoc. Based on harnessing the power of Machine Learning (ML) algorithms, security researchers are developing novel techniques to more effectively detect botnet activity. These approaches analyze vast troves of system data to discern subtle patterns obscured by human cognition alone. The Aposemat IoT-23 dataset provides a treasure trove of benign and malware-infected traffic captures gathered from diverse Internet of Things (IoT) landscapes. Within its records lay clues that machine learning models can extract, aiding relentless adaptation to emerging threats without constant human supervision. No strategy is foolproof, but by leveraging technology’s ability to process exceedingly large datasets with minimal bias, we can enhance our protections. Further refinement of these techniques will strengthen our defenses against disruptors who wield botnets as a weapon. This study determines the performance of various ML algorithms including Logistic Regression (LR), Multi-Layer Perceptron (MLP), Naïve Bayes (NB), Decision Trees (DT), and Random Forest (RF), and emphasizes the robustness and weaknesses of each classifier. The highest accuracy in detecting botnets can be found using Random Forest and Decision Trees. Multiple classifiers are combined together to develop a layered model approach to assess its potential to enhance detection effectiveness. Data quality, evolving threat landscapes, and computational resource requirements are a few key challenges which are discussed in our research. Key findings emphasize the importance of a robust and adaptive ML based model for significant botnet detection in the IoT landscape. Future researches should focus on developing more optimal methods of using layered model architecture and improve model robustness against adversarial attacks.

Edge computing represents a significant paradigm shift in distributed computing by moving computational processing away from centralized cloud infrastructures to the edge of the network, closer to data source in a decentralized fashion. This decentralized solution is designed primarily to cater to the needs of an increasing number of Internet of Things (IoT) and smart devices and thus reinforcing the capabilities of cloud computing by reducing its limitations in terms of intrinsic latency, bandwidth, and real-time responsiveness for a dedicated application. This chapter methodically examines the fundamentals of edge computing by exploring how cloud-based architectures have evolved into edge-enabled systems, highlighting the interactions between the cloud, edge nodes, and fog layers. The architectural analysis highlights the crucial communication channels and encryption protocols that allow for real-time processing and dissemination capabilities. The network flow between the fog layer, intermediate processing nodes, and edge devices to demonstrate the technologies of this paradigm along with other key technical elements are discussed to create intelligent and responsive systems. This chapter also presents numerous application areas like smart grids, autonomous cars, healthcare monitoring, and industrial IoT where edge computing plays a key role. It has been discussed that integrating edge computing with 5G networks and AI improves quality of service, guarantees a reliable low-latency communication, and maximises distributed intelligence at scale. Additionally, a critical assessment of the difficulties involved in this advanced paradigm is given, covering data governance, security and privacy issues, interoperability gaps, infrastructure heterogeneity, and regulatory compliance. In conclusion, edge computing complements cloud computing rather than replacing it, facilitating distributed ecosystems that are secure, scalable, and real-time. It will play a vital role in shaping the future digital infrastructures due to its high capacity to support data processing and decision-making at the millisecond level. It provides an advanced and promising architecture for uninterrupted research and innovative industrial applications.

Mental illnesses are a major public health concern and one of the main causes of disability around the globe. They also provide substantial social and economic challenges. The non-invasive, real-time capabilities of electroencephalography (EEG) have made it a potential technique for identifying neural patterns linked to a range of mental diseases. A systematic assessment of EEG-based learning models for mental disease identification is presented in this article, underscoring the expanding use of AI and machine learning in neurodiagnostic. In addition to reviewing existing literature, we implement and evaluate three fine-tuned learning models (FTLM-1, FTLM-2, and FTLM-3) alongside three widely used transfer learning models (VGG19, ResNet50, and Mobile net). Comparative analysis reveals that the fine-tuned models achieved higher classification accuracy (up to 75%) compared to transfer learning models (up to 65%), underscoring the potential of task-specific model optimization in EEG-based mental health diagnostics. This work contributes to the advancement of intelligent, data-driven approaches for early and accurate detection of mental disorders.