Journal of Computing & Biomedical Informatics

The Advanced AI Techniques for Deepfake Audio Detection

Mon, 01 Sep 2025 00:00:00 +0000

Sharing and retention of information is critical in the growth of the society especially in the current world of technology. As much as technology has led to the revolutionization of sharing knowledge and information, it has also come with challenges, like misinformation. A recent issue of concern is the very persuasive audio deepfakes, artificially created audio clips that are meant to sound like real people. This is highly threatening especially in the professions such as journalism and in the social media when reliability is highly valued. To resolve this problem, Developed Sonic Sleuth, a new tool to detect audio deepfakes. It is based on state-of-the-art deep learning approaches that are able to discriminate between authentic and synthetic audio correctly by means of a custom convolutional neural network (CNN). An elaborate dataset, ASVspoof 2021, which included real and synthetic audio was employed to perform an intensive test. The model was able to perform impressively with no less than 97.27 percent accuracy by incorporating the background noise and the diversity of language. The purposed model gives better accuracy as compared to existing model.

Optimal Reactive Power Dispatch with Electric Vehicle Charging Loads using an Artificial Protozoa Optimizer

Muhammad Bilal, Tahir Mahmood, Muhammad Munwar Iqbal — Mon, 01 Sep 2025 00:00:00 +0000

Minimizing real power loss in electrical power systems is crucial for economic efficiency and grid stability. The increasing number of EVs (electric cars) compounds this problem because EV charging demands are very unpredictable and put a lot of pressure on the power system. In such a setting, the already important Optimal Reactive Power Dispatching (ORPD) problem takes on an even greater significance. To address the ORPD issue while taking EV charging demands into account, this study suggests an Artificial Protozoa Optimizer (APO). For reduction of power losses and enhance the voltage stability, the algorithm sees the best position and size of shunt capacitors. On standard IEEE 14 bus, 30 bus, and 33 bus systems, the APO's effectiveness was tested under different EV load scenarios. Even when dealing with unpredictable electric vehicle loads, the results reveal that the IEEE-33 connection system, the IEEE 30 bus system, and the IEEE 14 bus system all considerably lessen the total real loss of power by a considerable margin: 7.49%, 0.79%, and 0.31%, respectively. Additionally, the optimization greatly enhanced and maintained within acceptable ranges the voltage profiles. In modern power systems that include a lot of EVs, APO is a powerful and efficient tool for addressing the problem of ORPD, according to this study.

A Predictive Model and Performance Evaluation in Mathematics for Primary Education

Marvi Abro, Imtiaz Husain, Syed M. Hassan Zaidi, Faryal Sheikh, Ghulam Murtaza — Mon, 01 Sep 2025 00:00:00 +0000

This study investigates the application of predictive modelling to assess and forecast students’ academic performance in primary mathematics education. Four regression techniques, Linear Regression, Decision Tree Regression, Random Forest Regression, and K-Nearest Neighbours Regression, were implemented and comparatively evaluated. Model performance was measured using Mean Squared Error (MSE) as the primary metric. Results indicate that Linear Regression achieved the lowest MSE (1.33), establishing a strong predictive baseline. Although Decision Tree Regression effectively captured non-linear patterns, it yielded a substantially higher MSE (62.38), highlighting the risk of overfitting. Random Forest Regression improved generalization by aggregating multiple decision trees, achieving an MSE of 25.21. Meanwhile, K-Nearest Neighbours Regression provided localized predictive accuracy with a competitive MSE of 19.29. Collectively, these findings contribute to the growing body of research on data-driven approaches in education, providing practical insights for educators and policymakers to leverage predictive analytics and enhance learning outcomes in primary mathematics.

A Hybrid Machine Learning Framework for Early Diagnosis of Alzheimer’s Disease

Mon, 01 Sep 2025 00:00:00 +0000

Alzheimer’s disease (AD) is neurodegenerative disease, making its initial detection essential for suitable treatment planning and for slowing cognitive decline. This research addresses the challenge of accurately identifying Alzheimer’s in its initial stages, where subtle structural and functional brain changes often remain undetected through conventional diagnostic methods. Despite advancements in diagnostic technologies, a significant gap still exists in achieving both high accuracy and reliability for early detection. To bridge this gap, we propose a hybrid machine learning approach that integrates CNNs used feature extraction with a Support Vector Machine classifier of final decision-making. CNN effectively extracts discriminative spatial features from MRI and fMRI scans, while the SVM enhances classification by refining decision boundaries. The performance of Models was evaluated using metrics such as accuracy, precision, recall, and F1-score. Experimental results show that the proposed framework achieves 94.2% accuracy, 96% precision, 95% recall, and a 95.5% F1-score, outperforming conventional standalone techniques. These conclusions highlight the robustness the hybrid model for capturing complex patterns for reliable early diagnosis of AD. Furthermore, this study presents an efficient diagnostic tool that can support clinicians in timely interventions. Future research may extend this work by integrating multimodal data to further enhance predictive performance.

Predictive Analysis on Project Management Success through AI

Muhammad Hamid Qureshi, Muhammad Usman Sattar — Mon, 01 Sep 2025 00:00:00 +0000

The modern business landscape is evolving rapidly, and each project comes with its own set of challenges and complexities. This calls for new and more inventive methods of handling such projects, as this area of work is becoming increasingly intricate and fluid. This study is centered on the predictive use of AI (Artificial Intelligence) technologies like ML (Machine Learning) and LLMs (Large Language Models) to ensure more effective project management at each of the ten PMBOK® knowledge areas. Merging the qualitative feedback from senior project managers and the quantitative KPIs—budget variance, schedule adherence, stakeholder satisfaction, and risk response time—from 84 organizations in construction, pharma, IT, finance, and manufacturing based in the EU, UK, USA, and Middle East provides richer insights. The analysis demonstrates how advanced AI tools, from predictive analytics to intelligent chatbots, streamline a project’s life cycle by enhancing efficiency, acuity, and overall decision-making. Predictive AI is demonstrated to bolster schedule and risk management as well as stakeholder interaction. Traditional metrics such as schedule creation and risk detection indicate a significant improvement for AI-supported projects, with 50% and 25% improvement respectively, as well as 30% higher stakeholder satisfaction.

A Hybrid Intrusion Detection System for Security of Edge-Based IIoT

Nimra Sajjad Hussain, Amna Sattar, Mariam Fiaz, Hira Mustafa, Zainab, Soha Ali — Mon, 01 Sep 2025 00:00:00 +0000

Emergence of cloud computing and IoT technology in healthcare, telecommunications and Industry 4.0 (IIoT), has revolutionized most of the daily services. But this development has also made the aspect of security to be more advanced and complicated. IIoT system security is one of the primary concerns of any industry and researchers. IDS have also materialized as a key part of identifying malicious activity and in attempts to further enhance the security of the IIoT networks. IDS are highly adopted in detecting the real time attacks and making secure decisions. This study proposes a machine learning base intrusion detection system comprises of PCA and XGBoost for edge-based IIoT. The framework combines the techniques of misuse and anomaly detection. It employs Principal Component Analysis (PCA) for dimensionality reduction of the features and Extreme Gradient Boosting (XGBoost) as the intrinsic classifier. PCA makes training faster whereas XGBoost makes detection more accurate The system is evaluated using NSL-KDD and Bot-IoT benchmarks. On NSL-KDD, It obtained detection rate of 98.5%, accuracy of 99.2% and false alarm of 2.6%. It recorded 98.3% accuracy, 97.7% detection rate and 2.8 % false alarm rate on Bot-IoT. These findings indicate that the suggested framework is superior to current IDS models.

Evaluating Machine Learning -Based Intrusion Detection in Software Defined Networks Using NSL-KDD Dataset

Mon, 01 Sep 2025 00:00:00 +0000

Software Defined network is an emerging and evolving network technology. A significant advantage of SDN is that it offers centralized control of the network, where all controller operations are centralized. The open-source Software Defined Networking (SDN) emulator, Mininet, has been utilized for generating and simulating virtual networks, in conjunction with POX, an open-source remote controller. A publicly available dataset, NSL KDD, is utilized for intrusion detection, evaluation, and comparison among several classification algorithms. In this proposed work, a supervised machine learning algorithm, a Decision Tree, is utilized for intrusion detection. Python is used as a tool to create, test, and compare different approaches to detect malicious attacks and identify the best one among them. These experiments are conducted to achieve results based on accuracy, recall, false positives, F-measure, and precision. Our proposed ML approach possesses more potential for intrusion detection using an accuracy measure. The decision tree is the best approach to detect intrusion based on detection speed and effectiveness. The precision of a decision tree is also the highest and most accurate among machine learning techniques.

A Multidomain Virtual Framework for Sacral Neuromodulation: Integration of CT-Based Anatomical Modeling, Electrode Placement Optimization, and Closed-Loop Device Simulation

Muzamil Ahmed, Sarah Tariq, Tooba khan, Gul Munir, Saeed Ahmed, Natasha Mukhtiar — Mon, 01 Sep 2025 00:00:00 +0000

Sacral neuromodulation (SNM) has emerged as an effective third-line therapy for overactive bladder (OAB), fecal incontinence (FI), neurogenic lower urinary tract dysfunction (NLUTD), and nonobstructive urinary retention. However, challenges remain in lead placement accuracy, stimulation efficiency, and device longevity. In this work, we present a comprehensive virtual framework that integrates medical imaging, 3D anatomical modeling, Multiphysics simulation, and system-level instrumentation to optimize SNM therapy. A pelvic CT scan was segmented using 3D Slicer to reconstruct patient-specific anatomy of the sacral plexus and bladder. The reconstructed model was imported into Fusion 360, where realistic 3D geometries were developed and five distinct electrode placements were virtually designed. COMSOL Multiphysics was employed to analyze electric field distribution, current density, and activation zones, enabling objective quantification of placement efficacy. Additionally, a complete instrumentation framework was simulated, including wireless power transfer, microcontroller-based stimulation control, rectification, and closed-loop feedback from bladder sensors. Results indicated that electrode placements within 3 mm of the sacral plexus and an insertion angle of 35–40° achieved superior response scores and minimized revision risk. The integration of anatomical modeling with device-level circuit simulation highlights a pathway toward patient-specific, adaptive, and energy-efficient neuromodulation. This multi-domain approach enhances the translational potential of SNM, offering insights into both clinical efficacy and engineering feasibility.

Privacy-Aware E-Health Data Sharing via Decentralized Blockchain System

Mon, 01 Sep 2025 00:00:00 +0000

With the growing sensitivity of personal health information (PHI), ensuring secure and privacy-preserving mechanisms for data exchange has become a critical challenge. Blockchain technology, with its inherent properties of immutability, decentralization, and transparency, shows significant promise in reshaping healthcare data management. Unlike existing single-layer approaches, this paper presents a decentralized blockchain-based multi-layer architecture with trap-door based searchable encryption, designed to enable secure, scalable, and privacy-aware sharing of electronic health records (EHR), consisting of data generation, storage, service, and super service layers. It leverages private and consortium blockchains to preserve data confidentiality and enforce access control through smart contracts. Trapdoor-based searchable encryption enables privacy-preserving queries on encrypted records, ensuring sensitive PHI remains protected yet discoverable by authorized users. Experimental evaluation demonstrates improved access efficiency, reduced cost, and compliance with data privacy regulations. This work highlights blockchain’s transformative role in healthcare by ensuring trust, security, and accessibility.

BBR v3-Cubic smackdown: A Fairness and Convergence Quantitative Evaluation

Muhammad Ahsan, Sadia Abbas Shah, Farrukh Nadeem, Raybal Akhtar — Mon, 01 Sep 2025 00:00:00 +0000

Congestion control has been an important component of TCP for nearly four decades now. Bottleneck Bandwidth and Round-Trip Time (BBR) has brought a major change in the ways the congestion onset can be monitored, and proactive measures can be taken instead of a reactive technique being used in loss-based traditional algorithms, such as Cubic and Reno, for a long time. With the introduction of the latest iteration of BBR, BBR v3, it has been claimed that it has improved its co-existence with Cubic flows with better fairness, but convergence issues within BBR-v3 have been reported. No existing study quantifies BBR v3's fairness with Cubic, and the convergence of BBR v3 for intra-protocol streams needs to be quantified for each stream’s throughput as well. The simulations and emulations generally don’t bring the true picture of the performance of a congestion control algorithm. In this paper, we have evaluated BBR v3 with cubic using our real-time physical testbed using Jain’s Fairness Index to bring a more accurate fairness analysis of BBR v3 and Cubic streams. For convergence, a well-established statistical metric that measures the relative stability of throughput, known as the Coefficient of Variation (CoV), has been calculated for BBR v3/Cubic flows. We used Flent (a FLExible Network Tester) to perform rigorous tests using various pairs of streams in upload, and the results, along with the metadata, have been saved for reproducibility and validation. Our thorough testing on both wired (Ethernet) and wireless (Wi-Fi 4) testbeds confirms that fairness issues between BBR v3 and Cubic streams persist. These issues are particularly serious as the number of streams increases. Similarly, our convergence tests confirm that BBR v3 flows, especially the first stream, obtain a larger share of the bandwidth, and the throughput of each stream remains highly volatile.

Preserving Critical Signals in Magnetic Image Denoising: A Deep Learning Approach with Selective Feature Preservation

Muhammad Umair, Khalid Hamid — Mon, 01 Sep 2025 00:00:00 +0000

Noise contamination is a major issue in medical imaging because it affects the clarity of structures and can impact how doctors make diagnoses. To address this, this study introduces a new deep learning method called DenoiseNet. The main goal is to reduce noise without losing important details of the body’s anatomy, which is a challenge with traditional filtering techniques and standard CNN models that often smooth out too much and lose key information. DenoiseNet builds on the U-Net structure by adding spatial attention, channel attention, and residual blocks. These components help the model focus on noisy areas, highlight important features, and ensure that the learning process works smoothly. The model uses residual-attention fusion in the bottleneck, extracts important features in the encoder, and restores clear images in the decoder using skip connections and residual attention blocks. A hybrid loss function that combines MSE and SSIM helps balance pixel accuracy with how realistic the image looks, improving both noise reduction and structure preservation. Hybrid DenoiseNet, incorporating spatial and channel attention along with residual U-Net blocks, achieves a PSNR of 32.27 dB and SSIM of 0.9598. The performance is robust in both our Salt & Pepper noise dataset as well as a semi-synthetic MRI dataset—outperforming both BM3D (31.9 dB, 0.9862) and DnCNN (31.5 dB, 0.8826) under identical test conditions. These qualitative gains are a demonstration of improved noise suppression without loss of structural detail. This approach's strength is its capacity to produce encouraging outcomes even in the early phases of training, exhibiting consistent performance and the possibility of more gains with more time spent training. In comparison to conventional methods, the model gains improved feature representation and better convergence by including attention and residual learning into the U-Net backbone. When taking everything into account, the proposed DenoiseNet demonstrates that merging residual learning with attention mechanisms on a U-Net structure creates a powerful and effective approach for removing noise from medical images. The results show that the model preserves key anatomical details essential for accurate clinical analysis while also effectively reducing noise. These outcomes highlight DenoiseNet's potential as a robust framework that can be further improved and adapted for different types of medical imaging, paving the way for better patient outcomes and more reliable diagnoses.

The Significance of Blockchain Technology in Preserving Digital Evidence Integrity in Forensic Investigations

Marwa Popal, Ashfaq Ahmad, Muhammad Arshad — Mon, 01 Sep 2025 00:00:00 +0000

Digital forensics is the application of technology to the investigation and accounting of digital crime and the extraction and analysis of digital evidence from systems and networks. As the number of IoT (Internet of Things) devices has proliferated, the complexity inherent in forensic investigations has skyrocketed. While IoT forensics addresses the challenges of a lack of standardized security protocol and ensuring the integrity as well as security of evidence data, it brings to bear all the difficulties existing in classic forensics. Unfortunately, traditional digital forensic tools are generally inadequate in shielding that evidence from tampering or unauthorized access. Digital evidence in IoT environments has been viewed as a potential solution that blockchain technology can offer with its tamper-proof and immune characteristics. This research explores how permissioned blockchain frameworks can enhance the integrity, confidentiality, and traceability of digital evidence in IoT forensic investigations. It hypothesizes that blockchain-based architectures outperform traditional evidence-handling mechanisms in preserving evidentiary reliability. While most encouraging, key findings show blockchain can do this establish data integrity, reliability, and confidentiality while featuring a very strong chain of custody for evidence. Furthermore, the study proposes how blockchain can be added to forensic practices to improve security and trust in forensic investigations. This study holds implications not only for digital forensic professionals but also for stakeholders in sectors such as healthcare, smart city governance, and critical infrastructure, where the reliability of digital evidence is paramount. By demonstrating blockchain’s role in safeguarding forensic data, this research contributes to trust-building mechanisms essential for digital justice and regulatory compliance in cross-border investigations.

Lung Cancer Classification through Transfer Learning and Deep Feature Extraction using EfficientNetB3

Mon, 01 Sep 2025 00:00:00 +0000

Worldwide, lung cancer is one of the deadliest diseases. It is critical to make an early diagnosis of lung cancer for treatment. The standard process of diagnosis by a pathologist is by examining the histopathology images. The assessment of images by a pathologist is still prone to errors and time-consuming. To enhance the speed of the entire process and to accurately diagnose cancer in images, an automated procedure of cancer diagnosis is essential. In this paper, an automated process for lung cancer classification is proposed by using a pretrained deep model, EfficientNetB3. The proposed scheme uses lung cancer histopathological images, which are obtained from the ‘LC25000’ dataset. During the preprocessing phase, images were resized to a fixed dimension of 224x224x3 pixels. EfficientNetB3 was trained over 15,000 images of lung cancer classes, including squamous cell carcinoma, adenocarcinoma, and benign. After transfer learning, features are extracted from the model's second-to-last layer, or the dropout layer, which has an Nx512 dimension. The classifiers are then fed the feature vector as input to classify lung cancer after it has been divided into 80%, 10%, and 10% for training, validation, and testing. The proposed methodology is evaluated by different performance metrics, i.e., accuracy, recall, precision, FNR, TNR, FPR, F1-score, and misclassification rate. The results show that achieved the highest accuracy of 0.9980 with a misclassification rate of 0.0020, which was achieved by the proposed technique for Bilayered NN, which represents an improvement over the state-of-the-art techniques.

A Convolutional Neural Network and Vision Transformer Based Framework for Effective Detection of Liver Cancer

Mon, 01 Sep 2025 00:00:00 +0000

Liver cancer, particularly hepatocellular carcinoma (HCC), remains one of the most prevalent and lethal malignancies worldwide, underscoring the urgent need for early and reliable diagnostic solutions. Conventional diagnostic methods using computed tomography (CT) imaging are often limited by inter-observer variability and the high cognitive burden on radiologists. To address these challenges, this study proposes a hybrid deep learning framework that leverages Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for effective liver cancer detection. The research employs the publicly available 3D-IRCADb1 dataset of contrast-enhanced CT scans, with preprocessing and augmentation techniques applied to enhance model generalization. Three state-of-the-art architectures, EfficientNet-B0, TinyViT, and MobileViT v2, were trained and evaluated to assess their diagnostic performance. Among these, MobileViT v2 demonstrated superior performance and efficiency in classification tasks. To enhance clinical trust, Gradient-weighted Class Activation Mapping (Grad-CAM) was integrated to provide visual explanations of model predictions, highlighting regions of interest corresponding to tumor areas. The findings indicate that the proposed framework not only ensures robust diagnostic capability but also introduces interpretability and efficiency, making it suitable for deployment in clinical and resource-constrained environments. This research contributes to advancing AI-driven liver cancer diagnostics by bridging the gap between performance and transparency, ultimately supporting earlier detection and improved patient outcomes.

Novel Approach to Detect Verticillium Wilt Using Transfer Learning

Hamza Naveed, Tipu Sultan Ali Shah, Junaid Asghar, Talia Noreen, Atiqa Safdar — Mon, 01 Sep 2025 00:00:00 +0000

The detection of Verticillium wilt at an early and accurate stage is crucial for successful disease control measures. Aside from causing large cop losses worldwide year after year, this disease is very common in Taiwan. Standard laboratory detection techniques, such as pathogen isolation and serological testing, are time-consuming, labor-intensive, and require professional expertise. Deep learning methods, especially convolutional neural networks (CNNs), have only become available in recent years and are now proving to be effective tools for disease detection using images. This holds whether they're being used as satellites monitoring agriculture fields or by farmers keeping close tabs on their crops. Among these approaches is transfer learning -one branch of deep learning-uses pre-trained models to speed the development process and improve image classification. The goals of this research are to establish a new, precise, and frugal way of detecting Verticillium wilt in cotton plants by using transfer learning based on pre-trained deep networks (like ResNets or convolutional neural networks). The results of this research provide a significant contribution to the current discussion on using sophisticated convolutional neural network architectures for image classification applications. The results offer a road map for scholars and professionals who want to choose models that strike a compromise between precision, computational effectiveness, and generalization. It is possible to utilize several ready-to-hand image databases. Furthermore, the system attempts to address some of the traditional detection methods 'shortcomings with earlier and less invasive diagnosis to improve disease management and sustainable cotton production.