AI And Cyber Security In Cyber – Physical Systems – Atul Mishra (1)

4 DNN model architecture Model Architecture (See Fig. 4). Model Compilation The following setup was used to compile the model: 1. Binary cross-entropy is a loss function that works well for binary classification problems. 2. Adam, an adaptive learning rate optimiser, strikes a compromise between accuracy and speed when training. 3. Metrics: Accuracy, which tracks the percentage of accurate forecasts. Training and Evaluation After class imbalance treatment strategies used, the balanced dataset was divided into train and test sets.

We trained our model on the training set to ensure strong performance in anomaly detection in terms of accuracy, precision, recall, F1 score and AUC-ROC (Table 5 shows the Results). Table 5 Performance of different data balancing techniques Sampling technique Accuracy Precision Recall F-1 score 0.99 1.0 0.98 0.98 Random under sampling 0.55 0.032 1.0 0.063 Class weight adjustments 0.9851 0.0 0.0 0.0 As depicted in Fig.

5, the model’s training accuracy consistently increases, approaching 98% after 15 epochs. The validation accuracy follows a similar trend, stabilizing around 97%. The minimal gap between training and validation accuracy indicates effective learning and generalization, with no significant overfitting observed. Fig. 5 Train versus validation accuracy plot for SMOTE Results: Fig. 6 shows the confusion matrix for the SMOTE technique. Fig. 6 Confusion matrix for SMOTE technique The following table summarizes the performance of the models on data balanced using each technique.

Precision: Percentage of all projected anomalies that were accurately identified. Recall (Sensitivity): Percentage of anomalies that were accurately recognized out of all actual anomalies. F1-Score: The model’s total capacity to identify anomalies is represented by the harmonic mean of precision and recall. AUC-ROC: The model’s capacity to differentiate between typical and unusual processes.

3.4 Interpretation of Results To sum up, we investigated three methods to handle class imbalance in the given scenario: SMOTE, Random Under-sampling, and Class Weight Adjustments. Even though SMOTE had the best accuracy (0.99) and best precision (1.0) suggests that it is best for accurately detecting abnormalities. At the expense of decreased overall performance, as seen by its low precision (0.032) and F1-score (0.063), Random Under-sampling performed exceptionally well in recall (1.0), guaranteeing that all abnormalities were recognized. Class Weight Adjustments failed to detect anomalies successfully since both precision and recall were zero, even though they maintained high accuracy (0.9851) and AUC-ROC (0.993).

These findings show how difficult it is to deal with class imbalance in real-world datasets and emphasize the necessity of carefully choosing, adjusting, or combining different approaches depending on the demands of the task. 4 Future Directions and Challenge 4.1 Adversarial Attacks on CPS Recent research shows that ML based anomaly detectors for CPS are vulnerable to adversarial attacks. These attacks create noise to deceive both rule checkers and neural networks, and down detection accuracy [28] by huge amounts.

The aim of this book series is to present state of the art studies, research and best engineering practices, real-world applications and real-world case studies for the risks, security, and reliability of critical infrastructure systems and Cyber-Physical Systems. Volumes of this book series will cover modelling, analysis, frameworks, digital twin simulations of risks, failures and vulnerabilities of cyber critical infrastructures as well as will provide ICT approaches to ensure protection and avoid disruption of vital fields such as economy, utility supplies networks, telecommunications, transports, etc.

in the everyday life of citizens. The intertwine of cyber and real nature of critical infrastructures will be analyzed and challenges of risks, security, and reliability of critical infrastructure systems will be revealed. Computational intelligence provided by sensing and processing through the whole spectrum of Cloud-to-thing continuum technologies will be the basis for real-time detection of risks, threats, anomalies, etc.

in cyber critical infrastructures and will prompt for human and automated protection actions. Finally, studies and recommendations to policy makers, managers, local and governmental administrations and global international organizations will be sought. 1kitap1.com/en Editors Atul Mishra, Pradeep Kumar Arya, Arvind Keprate and Alok Mishra AI and Cyber Security in Cyber-Physical Systems 1kitap1.com/en Editors Atul Mishra School of Engineering and Technology, BML Munjal University, Gurugram, Haryana, India Pradeep Kumar Arya Bennett University, Greater Noida, Uttar Pradesh, India Arvind Keprate Department of Mechanical, Electrical and Chemical Engineering, Oslo Metropolitan University, Oslo, Norway Alok Mishra Faculty of Engineeing, Norwegian University of Science and Technology (NTNU), Trondheim, Norway ISSN 2731-5002 e-ISSN 2731-5010 Engineering Cyber-Physical Systems and Critical Infrastructures ISBN 978-3-032-09655-5 e-ISBN 978-3-032-09656-2 https://doi.org/10.1007/978-3-032-09656-2 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2026 This work is subject to copyright.

All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

This is a short excerpt from the opening of “” by Unknown, quoted for review and introduction purposes. All rights belong to the copyright holders.

Book Information

• Unique ID: b9a3f93aae135ad0
• File Extension: .pdf
• File Size: 26,862,007 bytes (25.618 MB)
• Title: –
• Author: Unknown
• ISBN: 9783032096555, 9783032096562, 9783319758800
• Pages: 338
• Language: English (en)

Reading & Word Statistics

• Estimated Reading Time: 420.94 minutes
• Total Words: 84,189
• Total Characters: 598,627
• Average Words per Page: 249.08
• Average Characters per Page: 1771.09

Most Frequent Words

cps (1232), systems (911), data (774), security (656), system (489), detection (397), physical (352), learning (333), cyber (298), communication (293), attacks (275), control (263), cyber-physical (233), anomaly (229), smart (222), access (222), devices (195), threats (191), https (187), network (182), critical (180), org (174), challenges (172), machine (169), intelligence (163), employee (162), applications (156), real-time (154), healthcare (152), models (149), doi (148), industrial (147), attack (147), autonomous (140), artificial (138), ieee (138), anomalies (137), threat (137), secure (137), information (134), engagement (133), use (131), networks (130), using (127), techniques (126), crossref (125), also (123), example (123), energy (123), sensors (120), components (118), like (117), used (115), between (115), algorithms (113), sensor (106), model (106), new (105), analysis (103), processes (101), protocols (101), based (100), potential (98), traffic (98), safety (96), technologies (94), privacy (94), case (91), integration (91), fig (91), vulnerabilities (90), technology (90), risk (89), monitoring (87), one (86), issues (86), management (83), patterns (83), impact (82), methods (82), ai-driven (81), complex (81), layer (80), study (80), grids (79), integrity (79), key (79), cybersecurity (78), including (78), vehicles (77), medical (77), infrastructure (76), time (76), iot (76), reliability (75), digital (75), performance (75), predictive (75), various (75), manufacturing (74).

PDF Download