XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE Evaluate Hybrid Classical-Quantum Architecture in Deep Image Processing Abstract—This research focuses on the architectural design and the performance difference of hybrid quantum and classical machine learning in the context of image processing using Convolutional Neural Networks (CNNs). As we are in the digital era, it’s crucial to deal with complex datasets due to the scalability and performance issues. In this research, researchers experienced the use of classical CNNs integration with the quantum computing paradigms. Based on the research findings, we outline the architecture of such hybrid models, highlighting the integration of standard CNN layers with Variational Quantum Circuits (VQCs) and the prominence of the ReLU activation function. The study investigates comparative performance metrics and observations using standard CNNs and adding extra layers in contrast to those of quantum CNNs run on classical hardware, highlighting such metrics as accuracy and computational performance. The findings highlight the feasibility and comparative advantage of hybrid models, suggesting the prospects of improved sophisticated deep image processing methods. Keywords—machine learning, quantum computing, quantum machine learning, deep learning, image processing I. INTRODUCTION The exponential increase in the volume and complexity of data poses huge challenges in scaling up storage of the data and processing, as well as in increasing the efficacy of Machine Learning (ML) and Data Analysis (DA) methods [1]. Traditional ML methods, designed in the early days of smaller datasets and classical computing environments, are difficult to handle larger datasets because of their fundamental inability to utilize parallel processing and their limited processing speed [2]. In deep learning, the CNNs became very popular for the classification of images because of their skill in extracting features from the image in deep learning within machine learning. However, when dealing with larger datasets, the CNNs necessitate high processing power, making them less feasible for real-time in large-scale applications [3]. There is a significant impact on deep image processing in the fast-growing field of machine learning that offers a promising substitute for the traditionally used computational techniques. The classical Convolutional Neural Networks are increasingly limited in their scalability and computational power when dealing with large-sized and complicated datasets [4]. In the meantime, principles of quantum computing offer a viable solution for enhancing the level of accuracy, speed, and scalability in the analysis of data and computational velocity [5]. The paper explores the structural design of hybrid classicalquantum architecture by comparing its structure and functionality when undergoing simulations in classical hardware and deriving findings in the latest empirical research [6]. Although earlier research work has mentioned that Quantum CNNs (QCNNs) are able to yield high accuracy and computational efficaciousness than classical CNNs in the case of small datasets and binary classification problems, there is yet a significant research gap regarding their employment in the case of highly complex datasets [7], such as the Street View House Numbers (SVHN) dataset, which have significantly greater processing demands [8]. The paper has study to fill the gap through the detailing of the design and the simulated performance of the hybrid quantum-classical CNN structures. II. RELATED THEORIES OF HYBRID ARCHITECTURE A. Classical Machine Learning Machine Learning is a subset of artificial intelligence that includes the development of algorithms which can enable computers to learn from data and make decisions and predictions by analyzing patterns [9]. Deep Learning (DL) is one of the specific sub-fields of ML based on multi-layered artificial neural networks for learning end-to-end hierarchical representations from the data [10]. Unlike the majority of ML algorithms, whose feature extraction is human-engineered, the DL model can learn the features from the raw input; hence, it is highly successful for applications in unstructured data, such as image processing applications [11]. Convolutional Neural Networks are extensively used for deep image processing due to their high potential in feature extraction as well as pattern recognition in image information [12]. CNNs involve several layers, including the convolutional layer for feature detection, pooling for dimensional reduction, as well a fully connected layer for classification purposes [13]. Although CNNs are highly successful, the classical model is Niranjala S. H. Faculty of Information Technology City University Selangor, Malaysia shyama@gwu.ac.lk ORCID: 0009-0001-6580-4726 M. Kazem Chamran * Faculty of Information Technology City University Selangor, Malaysia * kazem.chamran@city.edu.my ORCID: 0000-0003-3836-4443 Mustafa Mowafak Alobaedy Faculty of Information Technology City University Selangor, Malaysia mustafa.theab@city.edu.my ORCID: 0000-0002-6562-2922 computationally expensive, particularly when using highdimensional image datasets in real-time applications [14]. Training of the networks is computationally expensive as well as memory-expensive, hence it can restrict the scalability as well as the implementation in real-time in the resource-constrained setting [15]. The rapid increase in complexity, as well as the size of image information, necessitates the design of more efficient models for the realization of fast computation without loss of precision [16]. Therefore, it’s essential to improve data analysis and deep learning using quantum computing technologies. B. Principles of Quantum Computing Quantum computing is an entirely new method based on the fundamental principles of quantum mechanics, employing the principles of superposition and entanglement in the performance of computational functions that significantly outpace classical computers [17]. The basic element of quantum computing is the qubit, unlike classical bits, which can be in several states at the same time, with superposition. Quantum computers with n qubits are then able to simulate 2 n states simultaneously, so they are able to deal with hundreds of billions of possibilities simultaneously [18]. Entanglement is an important quantum effect in which the quantum states of two or more qubits become fundamentally correlated, so that the state of a given qubit directly affects the state of another, irrespective of physical distance [19]. In Quantum Machine Learning, entanglement is of eminent importance for storing and processing complex correlations of the data, thus allowing the algorithms to unveil sophisticated patterns that could be missed when using classical techniques [20]. The states of the qubits are operated upon through the quantum gates, the analogs of the classical logic gates, but that operate on the principles of quantum mechanics, characterized through the unitary matrices. Single-qubit gates include the Bitflip, Pauli-Y Gate (bit and phase flip), phase-flip Pauli-Z Gate, and Hadamard Gate (superposition creator) [21]. The Controlled-NOT Gate or CNOT Gate is another basic multiqubit gate that makes the target qubit's state dependent on the control qubit and is required in the production of entangled states. The quantum circuits form the fundamental platform for the implementation of quantum computations, made up of a sequence of quantum gates and qubits, also in sequence [16]. The circuits are defined in terms of their acyclic form, such that the flow of the informational content is ensured to be unidirectional and complies with the no-cloning theorem, stating that the accurate duplication of all undefined quantum states is not possible. Measurement is the key probabilistic process via which we recover classical information from quantum states, so that superposition is reduced to a definite state [13,25]. C. Hybrid Classical-Quantum Architecture Hybrid quantum-classical machine learning protocols combine the best of the quantum as well as classical paradigms of computing [22]. For the case of deep image processing, this typically involves using a classical CNN-based front end for initial feature extraction, followed by a quantum back end employing quantum features for refined processing or classification [23]. This approach attempts to overcome the limitation of the current era’s availability of quantum hardware while taking advantage of the advantages provided by quantum computing for specific functions [14]. A crucial part of hybrid quantum convolutional neural networks (QCNNs) is the conversion of classical image data into quantum states, often achieved by using techniques like angle embedding. These quantum states undergo processing by Parameterized Quantum Circuits (PQCs) and Variational Quantum Circuits (VQCs) can be defined based on their variable parameters, which can be learned using classical optimization algorithms [19]. The circuit design of the QCNN is as a post-processing module, applying the operation of an inverse Quantum Fourier Transform (QFT) across qubits in order to act as a feature transformation and decoding layer [24]. This involves the implementation of several SWAP gates for qubit reordering, Hadamard (H) gate applications for the attainment of superposition, as well as controlled phase rotations (CP gates) for the creation of entanglement and phase information preservation [26,27]. This module can transform localized image features into holistic frequency-domain representations, similar to the application of the Fast Fourier Transform (FFT) in classical CNNs, enhancing translation invariance as well as feature generalization. Finally, the measurement values of the quantum circuit are passed back to the classical part for end classification as well as optimization. III. METHODOLOGY The study uses a systematic experimental design to analyze the applicability of quantum machine learning algorithms hybrid with conventional classical machine learning strategies for deep image processing tasks. This research design is exploring the prospect of quantum computing in increasing the efficiency as well as the effectiveness of machine learning mechanisms, with specific reference to the usage of CNNs for classification as well as evaluation of high-dimensional image datasets. The experimentation is set out in the form of an exhaustive sequence, gradually building phases to analyze performance indicators as well as computational advantage. Due to its characteristic features, the Google Street View House Number (SVHN) dataset has been picked as the baseline dataset as it is applicable for sophisticated image processing applications involving classic as well as quantum machine learning paradigms. SVHN contains real photographs of the house numbers taken from Google Street View, which feature much greater variability in image quality, lighting, viewpoint, scale, as well as in the distribution of colors when compared with simpler synthetic datasets like MNIST. This similarity with realistic scenarios makes it more suitable for ascertaining the performance and resilience of the models in scenarios, that closely mirror the application-use scenarios in reality. The SVHN dataset contains over 600,000 digit images, allowing for thorough evaluation of computational effectiveness as well as accuracy, necessary for the comparative analysis of classical as well as quantum-based CNNs. This dataset is split into a training subset of 73,257, a testing subset of 26,032, as well as a supplemental set of 531,131 of lower-quality images for additional training purposes. The moderate to high complexity of the SVHN dataset makes it particularly suitable for clearly illustrating potential quantum advantages in processing power as well as in speed of computation boosts. IV. MODEL DEVELOPMENT A. Classical Model Development and Training The first step in model development was the development and training of a classical CNN on classical hardware using standard machine learning techniques. The classical CNN provides the necessary baseline against which the performance of the quantum-enhanced models will be compared. The classical CNN architecture was constructed with multiple layers, including convolutional layers for feature extraction, pooling layers to reduce dimensionality, and fully connected layers for classification. The models examined specifically were the Baseline CNN and an Add Layer variant, both with the ReLU activation function. The models were trained on the 73,257image training subset of the SVHN dataset for 30 epochs, using the Adam optimizer and categorical cross-entropy loss. Throughout the training, important performance metrics such as accuracy, computational time, and resource utilization were carefully measured and recorded. B. Quantum Model Development and Training After the creation of the classical CNN baseline, the subsequent phase involved the conceptualization and simulation of the quantum-based machine learning model using classical computing platforms. This is the most crucial step for studying the theoretical merits as well as the practical viability of combining quantum algorithms with CNN architectures without the constraint of the capabilities of real quantum hardware. This QML model was proposed as one that combines quantum algorithms with the preestablished CNN model. Due to the present capabilities of quantum hardware, the performance of this QML model was simulated using classical computing platforms. Specifically, the quantum computing simulators as well as frameworks, for example, IBM’s Qiskit as well as Aer simulators, were adopted for the implementation of these simulations. Models simulated included the Baseline CNN as well as the Add Layer variant, using the ReLU activation function for the latter two. Models were trained as well as tested using the same SVHN dataset applied for the classical CNN, thus providing the same foundation for comparison purposes. The circuit design of the QCNN employed here was specially tailored to take advantage of quantum gates as well as their properties. QCNN circuit design applies the quantum gates, in this case, the inverse quantum Fourier transform (QFT) for four qubits, as part of the feature decoding layer of a four-qubit quantum machine learning state (q0, q1, q2, q3) with quantum-encoded images. SWAP gates re-order qubits after QFT, essential for recovering spatial or frequency structure after convolution. Qubits experience the Hadamard (H) process for putting qubits in superposition and controlled phase rotations (CP gates) for preservation of phase information for facilitating entanglement. This module expresses localized image features in the frequency domain descriptions, analogous to classical CNNs via the FFT, facilitating translation invariance as well as feature generalization. V. RESULT ANALYSIS The study focused on two types of experiments as follows. 1. QML using classical machine learning Classical CNN baseline models were constructed and meticulously tested to set the baseline performance. All the models were trained for 30 epochs using the SVHN dataset alongside the Adam optimizer. The "Baseline CNN" setting, as shown in Fig. 1, the ReLU activation function, obtained training accuracy as well as validation accuracy of 94%, while it had a validation loss of 18%. However, the "Add Layer" setting using ReLU showed considerable improvement, obtaining training accuracy of 97% as well as validation accuracy of 96%, with decreased validation loss of 11%. Fig. 1. Classical model: Baseline accuracy and loss Fig. 2. Classical model: Add layer accuracy and loss Among the ReLU-based models, the variant "Add Layer" had the best global performance, achieving Precision of 94%, Recall of 91%, and F1-score of 93%. The "Baseline CNN" achieved an F1-score of 89%, Precision of 90%, and Recall of 88%. The statistical representation illustrated in Fig. 2. This baseline classical CNN confusion matrix visually represented the classification accuracy for the whole set of the ten digit categories (0-9), with constant high values all along the main diagonal, which shows correct classifications (for example, for class 0 the correct classification is given by 4675, for class 1 by 3874). Off-diagonal elements signaled particular patterns of error of classification, for example, 136 occurrences in which true class 0 was misclassified as class 5, 97 occurrences in which true class 7 was classified as class 5, showing the specific regions in which the model was ambiguous in spite of the global excellent performance as shown in the Fig. 3 and Fig. 4 indicated the confusion matrix showing true house number prediction in Figure 1 Figure 2 “true table axis and the category that the model predicted for the image shows in “predicted label” using classical model. Fig. 3. Classical model: Precision and Recall per class Fig. 4. Classical model: Confusion matrix 2. QML using quantum simulated machine learning Hybrid quantum-classical models for the simulation of quantum learning behavior using classical processors have now started to appear at this juncture. They were based on the combination of the Variational Quantum Circuits (VQCs) in the CNN model using IBM's Qiskit software, along with Aer simulators for the simulation of quantum behavior. All the models were trained for 30 epochs using the SVHN dataset with the same image normalization, one-hot coding, and data splits as defined in the pipelines. Fig. 5. Hybrid Classical-quantum model: Baseline accuracy and loss Fig. 6. Hybrid Classical-quantum model: Add layer accuracy and loss Fig. 7. Hybrid Classical-quantum model: Precision and Recall per class For the model configurations using ReLU, "Baseline CNN" achieved 97% accuracy during training as well as in validation, training loss remaining at 9% and loss in validation at 11%. When there was the presence of an "Add Layer", the model improved, the training accuracy becoming 98% with the accuracy in the validation at 97%, training loss as well as validation loss dropping to 4% as well as 9% respectively as illustrated in Fig. 5 baseline statistics and Fig. 6 add layer’s statistical values. Among the model types applying Rectified Linear Units, "Baseline CNN" as well as "Add Layer" obtained the remarkable F1-score of 93% as shown in Fig. 7. "Baseline CNN" displayed strong performance indicators, whereby Precision was measured at 92% and Recall at 94%. Comparatively, "Add Layer" outperformed in all the statistical indicators, obtaining Precision as well as Recall of 93%. The confusion matrix created for the Fig. 8. Hybrid classical-quantum model: Confusion matrix simulated QML model revealed high effectiveness, marked by large values along the major diagonal, which indicate correct classification for the large part of the categories (4871 for class 0 and 3950 for class 1). Even with the extremely high accuracy, the off-diagonal terms expressed specific tendencies for the misclassification, for example, correct classification of true class 0 as class 3 (56 samples) or class 6 (43 samples), thus showing the specific uncertainty regions in the model, irrespective of the model's high performance in the whole as shown in Fig. 8, confusion matrix showing true house number prediction in “true table axis and the category that the model predicted for the image shows in “predicted label” in quantum-classical hybrid model. With respect to computational effectiveness, it was observed that hardware-level simulation of QML was exhaustive for the standard classical hardware. Nevertheless, the measured time of 130–152 ms per image was deemed suitable for the prototype execution, hence supporting the potential effectiveness of the QML simulation. VI. DISCUSSION This study evaluates and compares the effectiveness of different models using parameters such as accuracy, loss, precision, recall, and F1-score. The results are shown through the respective graphs and confusion matrices, which provide the trends of the performance for 30 epochs. Comparison of classical CNNs with quantum machine learning simulations run on classical hardware provides useful insights for the design as well as the performance of hybrid deep image processing designs. Baseline CNN, specifically the one using the Add Layer with ReLU activation, provided the best foundation, showing high raw accuracy along with high computational power. The "Add Layer" variant of the CNN was noticeably better compared to the "Baseline CNN" variant, showing the possibility of increased learning capability with increased depth, relevant for the analysis of existing classical notions of deep learning. Such analysis holds even for the case of quantum-aided designs, since these methodological strategies include the inherent factors of the optimization and generalization of the neural networks, relevant for the classical as well as the quantum setting. Simulated quantum machine learning models, using quantum gates, concepts, and principles on standard hardware, demonstrated the potential of incorporating quantum theories for creating a new abstraction level, allowing the identification of complex patterns. These models achieved competitive accuracy with classical CNNs, hence providing a viable prototype implementation. Both the "Baseline CNN" as well as the "Add Layer" versions, utilizing ReLU activation functions, demonstrated high performance in simulations, attaining high accuracy rates as well as high F1-scores. Although the hardware-level quantum simulation was computationally intensive for commonly available classical hardware, the 130– 152 milliseconds per image for recorded inference time was deemed tolerable for a prototype implementation, hence signifying the viability of quantum simulation. This "acceptable" inference time for simulated QML, despite its computational intensity. Simulation of quantum advantage can bring about considerable practical benefits, specifically in scenarios when classical CNNs can experience difficulties in fulfilling real-time applications. This suggestion implies that hybrid approaches can feature critically in filling the gap in the course of maturing fullscale quantum hardware, hence ratifying the need for simulating quantum models as the crucial developmental stage before full deployment on real quantum systems. The finding in the classical as well as simulated quantum phases, that architectural modifications impact model accuracy substantially regardless of the base computational paradigm, implies that the resulting classical deep learning strategies are anything but "classical," rather capture essential rules of neural network optimization and generalization cutting across the classical-quantum divide. This further means that the potential of quantum machine learning, at least in its current hybrid incarnations, lies in the combined power of quantum computation for tasks with powerful classical regularization and optimization protocols. VII. CONCLUSION This research provides an in-depth analysis of the design of hybrid quantum-classical CNN configurations for sophisticated image processing. This study systematically analyzed the combination of quantum computing concepts with classical CNNs simulations performed on classical hardware platforms. One of the contributions of this study is the empirical basis provided, showing the potential and competitive performance of these hybrid schemes, in particular, the ReLU-activated Baseline CNN and Add Layer variants, in dealing with highdimensional as well as intricate image data at epoch 30. These results highlight that the combination of quantum concepts can bring about a new level of abstraction, allowing for the identification of fine patterns as well as obtaining competitive accuracy for classical CNNs. Even the continued fundamental role of the increase in the number of layers, as well as the ReLU activation function in fine-tuning these hybrid configurations, was highlighted throughout this study. Even though existing hardware limitations of quantum hardware call for sophisticated simulation, the obtained performance shows the bright prospect of hybrid learning schemes for furthering deep image processing. Future studies will focus on refining these configurations further as well as studying implementation on increasingly powerful real quantum hardware platforms. REFERENCES [1] K. Abbas, M. Afaq, T. Ahmed Khan, and W.-C. Song, “A Blockchain and Machine Learning-Based Drug Supply Chain Management and Recommendation System for Smart Pharmaceutical Industry,” Electronics, vol. 9, no. 5, p. 852, May 2020, doi: 10.3390/electronics9050852. [2] G. Abdulsalam, S. Meshoul, and H. Shaiba, “Explainable Heart Disease Prediction Using Ensemble-Quantum Machine Learning Approach,” Intell. Autom. SOFT Comput., vol. 36, no. 1, pp. 761–779, 2023, doi: 10.32604/iasc.2023.032262. [3] F. Acheampong, “Big Data, Machine Learning and the BlockChain Technology: An Overview,” Int. J. Comput. Appl., vol. 180, pp. 1–4, Mar. 2018, doi: 10.5120/ijca2018916674. [4] R. Ahalya, U. Snekhalatha, and V. Dhanraj, “Automated segmentation and classification of hand thermal images in rheumatoid arthritis using machine learning algorithms: A comparison with quantum machine learning technique,” J. Therm. Biol., vol. 111, Jan. 2023, doi: 10.1016/j.jtherbio.2022.103404. [5] M. Akter et al., “Exploring the Vulnerabilities of Machine Learning and Quantum Machine Learning to Adversarial Attacks using a Malware Dataset: A Comparative Analysis,” in University System of Georgia, C. Ardagna, N. Atukorala, C. Chang, R. Chang, J. Fan, G. Fox, S. Helal, Z. Jin, Q. Lu, T. Seceleanu, and S. Yau, Eds., 2023, pp. 222–231. doi: 10.1109/SSE60056.2023.00037. [6] A. Ali et al., “Financial Fraud Detection Based on Machine Learning: A Systematic Literature Review,” Appl. Sci., vol. 12, no. 19, 2022, doi: 10.3390/app12199637. [7] J. Allcock and S. Zhang, “Quantum machine learning,” Natl. Sci. Rev., vol. 6, no. 1, pp. 26-+, Jan. 2019, doi: 10.1093/nsr/nwy149. [8] F. Amato et al., “QUANTUMOONLIGHT: A low-code platform to experiment with quantum machine learning,” SOFTWAREX, vol. 22, May 2023, doi: 10.1016/j.softx.2023.101399. [9] J. Amin, M. Sharif, N. Gul, S. Kadry, and C. Chakraborty, “Quantum Machine Learning Architecture for COVID-19 Classification Based on Synthetic Data Generation Using Conditional Adversarial Neural Network,” Cogn. Comput., vol. 14, no. 5, pp. 1677–1688, Sep. 2022, doi: 10.1007/s12559-021-09926-6. [10] K. Atz, C. Isert, M. Böcker, J. Jiménez-Luna, and G. Schneider, “Δ- Quantum machine-learning for medicinal chemistry,” Phys. Chem. Chem. Phys., vol. 24, no. 18, pp. 10775–10783, May 2022, doi: 10.1039/d2cp00834c. [11] E. Bahnsen, S. Rasmussen, N. Loft, and N. Zinner, “Application of the Diamond Gate in Quantum Fourier Transformations and Quantum Machine Learning,” Phys. Rev. Appl., vol. 17, no. 2, Feb. 2022, doi: 10.1103/PhysRevApplied.17.024053. [12] M. Broughton et al., “TensorFlow Quantum: A Software Framework for Quantum Machine Learning,” Aug. 26, 2021, arXiv: arXiv:2003.02989. doi: 10.48550/arXiv.2003.02989. [13] L. Buffoni and F. Caruso, “New trends in quantum machine learning,” EPL, vol. 132, no. 6, Dec. 2020, doi: 10.1209/0295-5075/132/60004. [14] O. Bouchmal, B. Cimoli, R. Stabile, J. Olmos, and I. Monroy, “From classical to quantum machine learning: survey on routing optimization in 6G software defined networking,” Front. Commun. Netw., vol. 4, Nov. 2023, doi: 10.3389/frcmn.2023.1220227. [15] T. Koike-Akino, P. Wang, Y. Wang, and IEEE, “AutoQML: Automated Quantum Machine Learning for Wi-Fi Integrated Sensing and Communications,” presented at the 2022 IEEE 12TH SENSOR ARRAY AND MULTICHANNEL SIGNAL PROCESSING WORKSHOP (SAM), 2022, pp. 360–364. doi: 10.1109/SAM53842.2022.9827846. [16] K. Bartkiewicz, C. Gneiting, A. Cernoch, K. Jiráková, K. Lemr, and F. Nori, “Experimental kernel-based quantum machine learning in finite feature space,” Sci. Rep., vol. 10, no. 1, Jul. 2020, doi: 10.1038/s41598- 020-68911-5. [17] R. Bhavsar et al., “Classification of Potentially Hazardous Asteroids Using Supervised Quantum Machine Learning,” IEEE ACCESS, vol. 11, pp. 75829–75848, 2023, doi: 10.1109/ACCESS.2023.3297498. [18] I. Petre, D. Iordache, M. Zamfir, M. Barbu, R. Duţescu, and I. A. Marinescu, “Virtual worlds, real technologies: an insight into Metaverse and its principles,” in 2023 24th International Conference on Control Systems and Computer Science (CSCS), May 2023, pp. 551–556. doi: 10.1109/CSCS59211.2023.00093. [19] M. Caro et al., “Generalization in quantum machine learning from few training data,” Nat. Commun., vol. 13, no. 1, Aug. 2022, doi: 10.1038/s41467-022-32550-3. [20] F. Amato et al., “QUANTUMOONLIGHT: A low-code platform to experiment with quantum machine learning,” SOFTWAREX, vol. 22, May 2023, doi: 10.1016/j.softx.2023.101399. [21] F. Cárdenas-López, M. Sanz, J. Retamal, and E. Solano, “Enhanced Quantum Synchronization via Quantum Machine Learning,” Adv. QUANTUM Technol., vol. 2, no. 7–8, Aug. 2019, doi: 10.1002/qute.201800076. [22] O. Ayoade, P. Rivas, and J. Orduz, “Artificial Intelligence Computing at the Quantum Level,” DATA, vol. 7, no. 3, Mar. 2022, doi: 10.3390/data7030028. [23] S. H. Niranjala, S. B. Goyal, and B. A. Kumar, “Secure and scalable data analysis framework with quantum machine learning and blockchain,” in Security Issues in Communication Devices, Networks and Computing Models, CRC Press, 2025. [24] S. H. Niranjala and S. B. Goyal, “Improving Wine Quality Forecasts Using Dynamic Integral Neural Networks and Optimized Against Interference,” Int. J. Smart Sustain. Intell. Comput., vol. 2, no. 1, pp. 52–64, Feb. 2025, doi: 10.63503/j.ijssic.2025.42. [25] H. Gupta, H. Varshney, T. Sharma, N. Pachauri, and O. Verma, “Comparative performance analysis of quantum machine learning with deep learning for diabetes prediction,” COMPLEX Intell. Syst., vol. 8, no. 4, pp. 3073–3087, Aug. 2022, doi: 10.1007/s40747-021-00398-7. [26] Kok-Lim Alvin Yau, Yung-Wey Chong, Xiumei Fan, Faranak Nejati, Mohammad Kazem Chamran, Shalini Darmaraju, Combinations of generative adversarial network and reinforcement learning: A survey, Neurocomputing, Volume 650, 2025, 130847, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2025.130847. [27] Niranjala, S.H., Alobaedy, M.M., Goyal, S.B. (2024). A Comparative Study of Machine Learning Techniques for Predicting Student Academic Performance. In: Vasant, P., et al. Intelligent Computing and Optimization. ICO 2023. Lecture Notes in Networks and Systems, vol 1167. Springer, Cham. https://doi.org/10.1007/978-3-031-73318-5_31