International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Quantifying Confidence in Diabetic Retinopathy Diagnosis: A Comparative XAI Study of Deep Learning and Bayesian Neural Networks
Get PDFDiabetic Retinopathy remains the primary microvascular complication of diabetes and a leading cause of irreversible blindness globally. While deep learning models offer high diagnostic accuracy, their widespread clinical integration is profoundly limited by two fundamental, unresolved deficiencies in previous literature: the absence of comprehensive, fair comparative analysis across diverse architectures and the pervasive lack of transparent, quantifiable prediction confidence necessary for clinical acceptance. This study directly addresses these challenges by presenting a highly optimized and rigorous comparative evaluation of three powerful models: the high-capacity EfficientNetB0, the computationally efficient MobileNetV3Small, and a novel Custom Bayesian Neural Network (BNN) framework. Through robust methodology, all models achieved exceptional generalization, stabilizing with impressive final F1-Score > 0.91. The Custom BNN demonstrated clear superiority as the most reliable diagnostic tool, securing the highest Accuracy 0.9294 and F1-score 0.9289 on the objective test set. Most significantly, this work delivers a breakthrough in safety assurance by integrating sophisticated Explainable AI (XAI) and probabilistic modeling: Grad-CAM and Local Interpretable Model-agnostic Explanations (LIME) confirmed anatomically grounded decision-making, while the BNN uniquely provides quantifiable uncertainty metrics, offering a crucial 95% confidence interval (CI) for every diagnosis. These results validate a new generation of high-performance models, led by a transparent BNN architecture, that are ready for implementation to deliver reliable, trusted, and efficient Diabetic Retinopathy screening solutions worldwide
Design and Validation of a Quantum Software Engineering Lifecycle Framework
Get PDFQuantum computing promises transformative computational capabilities; however, the absence of structured and standardized software engineering significantly limits its practical scalability. Quantum Software Engineering (QSE) remains fragmented, tool-centric, and largely experimental. This paper proposes a structured Quantum Software Engineering Lifecycle Framework that systematically integrates requirements engineering, hybrid design, quantum development, verification, deployment, and governance. To validate the proposed lifecycle, a two-stage evaluation was conducted. First, a Delphi-based expert validation involving 15 domain experts assessed clarity, feasibility, scalability, and hybrid applicability. Second, a simulation-based comparative analysis using Qiskit and PennyLane evaluated the lifecycle against ad-hoc development workflows across variational and search-based quantum algorithms. Results demonstrate a 24% reduction in development time, 15–18% improvement in execution fidelity, and significant gains in modularity, reusability, and tool interoperability. These findings confirm that adopting a structured lifecycle enables more reliable, scalable, and sustainable quantum software development, positioning QSE as a mature engineering discipline rather than purely experimental practice
AmpliStride: From Signal to Stride a Breakthrough for Leg Paralysis Rehabilitation
Get PDFFoot drop is a condition in which a patient fails to lift a foot due to neuro-muscular disorder of the lower body. Many assistive devices are available, but they have some limitations. For this purpose, we offer a solution: Muscle Signal Amplification and Transmission System (MSATS). The system picks muscle signal from the healthy leg and after necessary processing and amplification transmits it to the muscle stimulator worn on the affected leg. The system stimulates the muscle with proper timing according to the gait cycle. This project aims to improve the quality of life of those afflicted with foot-drop by assisting their mobility and independence. 
CHEESE Net: A Feature-Optimized Hybrid Learning Model
Get PDFIntelligent cheese selection is critical in the dairy industry to address rising consumer demand for personalized nutrition and health-conscious choices. This study introduces the novel integration of supervised learning, unsupervised clustering, and deep learning autoencoders to dynamically optimize feature representation and recommendation quality, a previously unaddressed approach in dairy informatics. The system employs Random Forest Regression for caloric prediction, PCA for dimensionality reduction, and deep autoencoders to capture non-linear nutrition relationships. Recommendations are generated via cosine similarity and Euclidean distance, supported by clustering techniques to refine cheese categories. Cheese net achieved exceptional predictive accuracy with a Mean Absolute Error (MAE) of 14.46 and an R² Score of 0.98, outperforming traditional models. Advanced visualizations (heatmaps, t-SNE, PCA plots) uncovered latent nutritional patterns while clustering enhanced recommendation precision by aligning suggestions with user-specific dietary profiles. The hybrid model’s interpretability enables stakeholders to decode correlations between fat, protein, carbohydrates, and moisture content, facilitating data-driven decisions for producers and consumers. By unifying machine learning with explainable AI, Cheese Net reduces MAE by 31% compared to standalone regression models. This framework pioneers a scalable, data-driven solution for personalized cheese selection, bridging nutritional science and consumer needs in the digital dairy era
Highly Isolated 4-Port UWB Mimo Antenna for Next Generation Communication System
Get PDFThis paper presents the design, optimization, and performance analysis of a compact four-port ultra-wideband (UWB) MIMO antenna for next-generation high-frequency communication systems. The antenna is built on a Rogers RT Duroid 5880 substrate and operates effectively in the 12.5–55 GHz, range making it suitable for millimeter-wave 5G applications. A four-step design process is used to develop a single antenna element optimized for wide bandwidth and good impedance matching. Parametric studies on feedline length, inset depth, and ground structure help improve bandwidth and ensure strong radiation patterns. In the MIMO setup, four radiating elements are placed at right angles to each other to reduce mutual coupling. Additionally, a centrally located plus- shaped decoupling resonator is added to further improve isolation, especially at lower frequencies, enhancing overall antenna performance. Simulation results show excellent impedance matching, a very low envelope correlation coefficient (ECC < 0.004), and a high diversity gain (dB). The antenna also delivers stable radiation patterns and high efficiency (>85 %) across the operating range. These findings confirm that the proposed MIMO antenna offers strong isolation (<-20dB), compact size (20x20 mm2), and wide bandwidth (40GHz) making it a suitable choice for future UWB and millimeter-wave MIMO systems
AI-Based Resource Efficient Image Classifier for Skin Lesions
Get PDFSkin cancer and other skin diseases are significant health concerns, and early diagnosis is essential for effective treatment. Traditional diagnostic methods, such as clinical examination and histopathological analysis, are time-consuming, require specialized expertise, and often cause delays in treatment. AI models have the potential to transform this process. While previous research has primarily focused on skin cancer or specific skin diseases, this study takes a broader approach by introducing a novel multiclass classification model. We created a unique dataset combining images from publicly available datasets and new images collected using mobile cameras. The dataset consists of three types of skin cancer and six categories of skin diseases, with both mobile camera and dermoscopic images included. In total, we gathered 6,820 skin lesion images, 4,957 from public datasets, and 1,863 new images to enhance the dataset. Various deep learning models, including VGG16, ResNet50, DenseNet121, MobileNet, and a custom CNN, were tested. While these models performed well with dermoscopy images, they struggled with mobile images. To address this, we implemented a new classification model, YOLOv11, for multiclass classification. This model achieved an impressive 97.5% overall accuracy, with an F1 score of 0.97503, and 99% accuracy for each class, handling both dermoscopy and mobile images effectively
Enhancing Driver Identification with a Crow Search-Optimized Stacking Ensemble
Get PDFDriver identification systems play a crucial role in enhancing vehicle security and delivering personalized experiences for drivers. Traditional identification methods typically use individual machine learning models, which often struggle to perform well due to their limited ability to adapt to diverse driving behaviors. In this study, we present a novel stacking ensemble framework optimized using the Crow Search Algorithm (CSA) to overcome these challenges. The CSA-optimized ensemble combines the strengths of several base models—Logistic Regression (LR), Naïve Bayes (NB), Random Forest (RF), and K-Nearest Neighbour (KNN)—with a meta-learner designed to boost both accuracy and robustness. CSA is used to fine-tune the ensemble’s hyperparameters, ensuring optimal performance. Experimental results on a driving dataset demonstrated that the proposed method significantly outperforms existing approaches in terms of identification accuracy, precision, and recall. This framework holds promise for a wide range of applications, including intelligent transportation systems and automotive cybersecurity
Challenges Faced by Stakeholders during the Requirement Engineering Phase: An Exploratory Study
Get PDFStakeholders are the backbone of any organization and play a vital role in the completion of any product. Different stakeholders with different roles, skills, natures, and experiences are involved throughout the Software Development Life Cycle (SDLC). Unlike other phases of SDLC, Requirement Engineering (RE) requires more stakeholders, active participation, focus, and collaboration. However, stakeholder involvement makes the RE phase more difficult and impacts other phases of Software Development. The inherent complexity of the RE phase is due to numerous factors, including diverse skill sets, language disparities, comprehension issues, and lack of interest, thereby rendering it particularly challenging for stakeholders. Literature also highlights some practices to resolve these issues, like enhancing communication and building trust among team members to overcome these challenges, but still, all these challenges affect software development in one way or another, and lead projects toward failure
Hybrid Deep Learning Approach for EEG-based Epilepsy Detection
Get PDFEpilepsy is a chronic neurological disorder characterized by continuous relentless seizures resulting from abnormal activity in the brain. Early and accurate diagnosis is very critical. The usual methods can take a lot of time for diagnosis and it can also often vary from one specialist to another. There have been many approaches implemented for detecting seizures with varying success. Electroencephalogram (EEG) analysis is a critical tool for diagnosing neurological conditions like epilepsy. A key focus in medical technology has been automating the detection of epilepsy but it has been challenging due to its complexity and large amount of data. Although the results of some studies have been encouraging, the use of these approaches has not been practical due to various issues i.e. imbalanced data signal variability to name a few. This research presents a new approach to improve performance and accuracy. A Hybrid Deep Learning model combines a number of paradigms of neural networks to leverage the best of multiple models in processing complex data like EEG signals. EEG. As EEG has both temporal and spatial data this hybrid approach is quite practical in handling different EEG components. In addition, a multimodal method is explored to enhance prediction performance. This involves enhancing EEG data with complementary data, such as clinical history and other biomarkers. Through integrating data from multiple sources, the model gains a broader context for epileptic activity detection. Which helps in bypassing the inefficiencies inherent in EEG signals. This combined approach can potentially provide stronger and clinically informative outcomes, hence enabling advancements in the early diagnosis of epilepsy
Sequestration of Carbon Dioxide via Mineral Carbonation to Produce Magnesium Carbonate: A Design Study
Get PDFThe rapid increase in atmospheric carbon dioxide (CO₂) due to industrialization and fossil fuel combustion has raised significant concerns about global warming. Carbon capture and storage (CCS) is a crucial technology for reducing greenhouse gas (GHG) emissions. This study presents the design of a mineral carbonation plant capable of sequestering 30 tons of CO₂ per day to produce magnesium carbonate (MgCO₃) using olivine as the feedstock.The process follows an ex-situ carbonation approach, where a mineral slurry reacts with CO₂ under controlled conditions. The plant design includes the development of key equipment such as a reactor, heat exchanger, and flash column, with a detailed process flow diagram (PFD) modeled in Aspen Plus. Material and energy balances ensure the operational feasibility of the system.With an effective conversion rate of 50%, the process accounts for realistic industrial limitations while maintaining reliability at scale. Heat recovery mechanisms, including a shell-and-tube heat exchanger, improve energy efficiency by minimizing heat loss. Optimized equipment design ensures process scalability and aligns with performance criteria to meet sequestration targets and product quality standards.The reliance on olivine, an abundant and cost-effective silicate mineral, highlights the economic and environmental advantages of this approach. The findings contribute to advancing sustainable CCS technologies, offering a viable solution for CO₂ mitigation while producing valuable industrial products such as MgCO₃ and by-product SiO₂