Robust Multimodal Fusion of Transfer Learning Framework for Leukemia Cancer Detection and Classification using Biomedical Images

Arwa Darwish Alzughaibi

doi:https://doi.org/10.54216/FPA.170230

Fusion: Practice and Applications | 2025

Fast Numeric Sign Detection Using Adaptive Thresholding and Geometry of Optimized Fingers

A strong sign language recognition system can break down the barriers that separate hearing and speaking members of society from speechless members. A novel fast recognition system with low computational cost for digital American Sign Language (ASL) is introduced in this research. Different image processing techniques are used to optimize and extract the shape of the hand fingers in each sign. The feature extraction stage includes a determination of the optimal threshold based on statistical bases and then recognizing the gap area in the zero sign and calculating the heights of each finger in the other digits. The classification stage depends on the gap area in the zero signs and the number of opened fingers in the other signs as well as the sequence in which the opened fingers appear for those that have the same number of opened fingers. The conducted test results showed the system’s high capability to classify all the digits; where both the precision and F-score percentages of the proposed model reached the desired optimal value (100%).

groups

Mela G. Abdul-Haleem mail -

Loay E. George mail

link https://doi.org/10.54216/FPA.190201

Volume & Issue

Vol. Volume 19 / Iss. Issue 2

Details open_in_new

Journal of Cybersecurity and Information Management | 2024

Proposal for the use of the neutrosophic method of Shannon entropy

The authors suggest a new approach to create a composite index by fusing of information as well as applying Shannon Entropy but ready under one challenge that exists when it comes to understanding the complex nature of the exercising data and that is integrating information from different sources that are distinct and provide different facets of an estimation into one single best. In that sense, the modern world of information system is one which has to work with data which is in most cases erroneous and sometimes even contradictory, then the perspective of fusing at such information in a fruitful manner becomes very important. However, such information is not always available as it is only after the uncertainty of the relevant data has been difficult to make optimal use of decision making ability of a model. This paper goes some way to assisting with that issue and describes a framework that involves multiple information sources as well as how information entropy models the data uncertainty. The method put forward in the study employs the integration of information fusion with the application of Shannon entropy computation to come up with a composite index that best represents the particular system of interest in terms of its intricacy and the amount of uncertainty associated with it. Furthermore, the application of this composite index has practical implications in various areas, such as risk management, business decision-making and public policy evaluation, where precision in information integration is crucial to achieve effective results.

groups

Leonso Dagoberto Torres Torres mail -

Milena Elizabeth Alvarez Tapia mail -

Paul Orlando Piray Rodríguez mail -

Aymuxammedova Amina Kakajanovna mail

link https://doi.org/10.54216/JCIM.140118

Volume & Issue

Vol. Volume 14 / Iss. Issue 1

Details open_in_new

Journal of Intelligent Systems and Internet of Things | 2025

IoT-Enabled monitoring system for Plant Health Growth

In current scenario, plant health monitoring plays a crucial role in effective health maintenance of plants in climate changes. Internet of Things (IoT) played an efficient role in realizing the remote and real-time monitoring of any physical things and activities through internet connectivity. In this study we have proposed a system that is able to monitor the plant health with the assimilation of wireless sensors and wireless network. The proposed system is able to log the sensor values on the plants on the cloud server through internet connectivity.

groups

Aditi Sharma mail -

Deepak S. Dharrao mail -

Kapil Joshi mail -

Vipin Tiwari mail -

Sumit Kumar mail -

Prabhat Kr. Srivastava mail -

Rahul Sharma mail

link https://doi.org/10.54216/JISIoT.140219

Volume & Issue

Vol. Volume 14 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2025

Revolutionizing E-Commerce Security: Unveiling an Innovative Deep Learning-Based Strategy for Detecting Financial Fraud

An inventive deep learning-based method for identifying financial fraud, revolutionizing e-commerce security in the process. The research offers a state-of-the-art setup that makes use of deep learning computations in the dynamic world of online exchanges, where the possibility of fraudulent activity is a danger. Since frauds are known to be erratic and lack consistency, it might be challenging to spot them. Con artists exploit the latest developments in technology. They manage to evade security measures, which results in millions of dollars being lost. One method of tracking fraudulent exchanges is to use information-mining techniques to investigate and detect unusual behaviours. Interactions. In contrast to deep learning techniques as auto encoders, convolutional neural networks (CNN), restricted Boltzmann machines (RBM), and deep belief networks (DBN), this paper aims to benchmark several machine-learning techniques, such as k-nearest neighbour (KNN), irregular forest, and support vector machines (SVM). The three-evaluation metrics that are really employed are the Area Under the ROC Curve (AUC), the Matthews Correlation Coefficient (MCC), and the Cost of Failure.

groups

Aditi Sharma mail -

S. Phani Praveen mail -

Vipin Tiwari mail -

Pradeep Kumar Arya mail -

Deepak Parvathaneni Naga Srinivasu mail -

Mukta Patel mail

link https://doi.org/10.54216/FPA.170227

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2024

Neutrosophic Methods and Linguistic Tools for Interpreting Human Perceptions in Complex Decision-Making

This study addresses a particular issue in relation to the disambiguation of human views, which remains critical in the current era that is the quest for suitable instruments that can cognize, simulate, and interpret the multilayered nature of the standpoint. Today, in contexts where a decision must be made that requires a synthesis of different and often-opposed points of view, such methods are very limited. This methodological gap focuses on the question where ways and means are lacking, which combine analytical accuracy and the flexibility of approaches for dealing with huge amounts of complex and unstructured information. To mitigate this problem, the study seeks for the application of neutrosophic methods and languages as a new approach for understanding human perceptions, which present a great deal of uncertainty. From the combined angles of neutrosophic logic and special linguistic devices, images from different practical situations are scrutinized. The results indicate that this method not only enhances the accuracy with which human subjectivity is simulated but also renders stronger analytical models for application in the area of organizational strategy, public policy formulation and even marketing research. In conclusion, this research extends new and significant methodological boundaries to the social and applied sciences and provides a useful approach to the problem of interpretation and decision-making in a multidimensional and time-changing society.

groups

María Lorena Merızalde Avıles mail -

Emver Nivela Ortega mail -

Kleber Eduardo Carrion Leon mail -

Wiem Abdelbaki mail

link https://doi.org/10.54216/FPA.160120

Volume & Issue

Vol. Volume 16 / Iss. Issue 1

Details open_in_new

Fusion: Practice and Applications | 2024

A Hybrid Neutrosophic Hierarchical Method with SWOT Analysis to Face Complexity and Uncertainty

This paper addresses the question that is global as decision making in the scenario of ambiguity. Given the conflicting or less dependable information, it also becomes necessary to look for approaches that assist us. Conventional strategic planning approaches work relatively well with straightforward and precise information. These become inadequate with situations that are ambiguous. To address this challenge, we adopt the Neutrosophic Hierarchy Method that integrates with SWOT analysis in addressing the challenge. As such, we learn to evaluate or assess the four components of SWOT: Strengths, opportunities, weaknesses and threats in wider terms. However, we do appreciate that often what we assess is not black and white but in shades of color. The conclusion is that for complex decision-making, this approach seems more appropriate and offers better results than others offer. The key aim of this article is to put forth a novel perspective on how decisions should be made in the face of uncertainty. Most of all, we expect to be helpful to both policymakers and strategists in the sense of providing a tool, which can be useful when it comes to the practical inconsistencies that are quite frequently in excess of reasonable solutions.

groups

Milena Avarez Tapia mail -

Carlos G. Rosero Martínez mail -

Josue R. Lımaıco Mına mail -

Saidkarimova Matlyuba Ishanovna mail

link https://doi.org/10.54216/FPA.160213

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Journal of Intelligent Systems and Internet of Things | 2025

Dynamic Feature Weighting for Efficient Multi-Script Identification Using YafNet: A Deep CNN Approach

Script identification is crucial for document analysis and optical character recognition (OCR). This study proposes YafNet, a novel convolutional neural network (CNN) architecture, developed from scratch, to tackle the challenges of script identification in both handwritten and printed word images. YafNet dynamically weights features, enabling it to learn and combine multimodal features for accurate script identification. To evaluate its efficacy, we use the imbalanced ICDAR 2021 Script Identification in the Wild (SIW 2021) competition dataset. Experimental results demonstrate that YafNet outperforms conventional approaches, particularly when trained on mixed handwritten and printed data. It achieves high classification accuracy, balanced accuracy, and ROC AUC scores, indicating its robustness and generalizability. The incorporation of data augmentation and external data further enhances performance, underscoring the model's potential for real-world applications.

groups

Yahia Menassel mail -

Rashiq Rafiq Marie mail -

Faycel Abbas mail -

Abdeljalil Gattal mail -

Mohammed Al-Sarem mail

link https://doi.org/10.54216/JISIoT.140220

Volume & Issue

Vol. Volume 14 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2025

Deep Learning-Based Steganalysis for Detection and Classification of Possible Hidden Content in Images

Steganalysis can be defined as the science that addresses the process of identifying and detecting hidden information or data within various types of digital media. Recently, Deep Learning (DL) approaches have been employed to build steganalysis systems. However, the problem with steganalysis systems adopting a DL approach is their low accuracy and their need for effective datasets to be used for the training. In this paper, we introduce a DL-based Steganalysis system for the detection and classification of hidden content in images. Our system, called Steg-Analysis Convolutional Neural Network (SA-CNN), relies on a Convolutional Neural Network (CNN) and uses High Pass Filter (HPF) and extra-embedded data. We also propose a preprocessing-based data hiding method to increase the accuracy of SA-CNN in detecting hidden content. Therefore, this ensures the imperceptibility of images used for training SA-CNN. In addition, we use another CNN, called Malicious-Benign Classification CNN (MBC-CNN), that we have developed to classify the extracted hidden content into Malicious or Benign classes. Compared with existing systems, SA-CNN shows a better performance in terms of accuracy, under increased hiding rates ranging from 0.1 to 1.0 bpp, reaching 90%.

groups

Mostafa A. Ahmad mail -

Eftkhar Al-Qhtani mail -

Ahmed H. Samak mail -

Amr Ibrahim mail -

Mourad Elloumi mail -

Ali Ahmed mail

link https://doi.org/10.54216/FPA.170228

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2025

Feature Selection based on Improved Differential Evolution (DE) Algorithm for E-mail Classification

Spam e-mail has become a pervasive nuisance in today's digital world, posing significant challenges to efficient communication and information dissemination. Dealing with huge amounts of data with irrelevant and redundant features, which leads to high dimension. Nowadays, with the growth of using the internet, finding the secure E-mail classification system for cloud computing is a very important topic. Additionally, determining the best algorithm for choosing a subset of features has a big impact on how well automatic email classification works, making it one of the major issues. Among these is the Differential Evolution (DE) algorithm, which is computationally costly because of the slow convergence rate and evolutionary process. To address these issues, this study offers an intelligent scheme called Opposition Differential Evolution (ODE), which combines the Opposition Based Learning (OBL) and DE algorithms for effective automated feature subset selection. Its effectiveness is assessed using the support vector machine (SVM) to present a strong performance when evaluating the e-mail spam classification rate. Moreover, the OBL is used to accelerate and increase the convergence rate of traditional DE. To determine which features, contribute most to the reliability of the email spam classification, the subset features based on ODE that was used as feature subset selection are used.To assess the effectiveness of the proposed scheme, extensive experiments are conducted on spambase” and “spamassassin” benchmark email datasets, comprising a diverse collection of spam and non-spam emails. The results demonstrate that the Opposition Differential Evolution (ODE) algorithm yields superior performance compared to traditional machine learning and evolutionary techniques, displaying its robustness and efficiency in identifying spam emails accurately. The ODE algorithm effectively handles high-dimensional feature spaces, enhancing the model's discriminatory power while maintaining computational efficiency. Compared to the suggested ODE-SVM technique, which yields a result of 96.79 percent, the full-feature accuracy result was 93.55 percent. Additionally, empirical results demonstrate that our scheme may efficiently increase the number of features needed to improve the accuracy of the email spam classification.

groups

Nadir Omer mail

link https://doi.org/10.54216/FPA.170229

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2025

Robust Multimodal Fusion of Transfer Learning Framework for Leukemia Cancer Detection and Classification using Biomedical Images

Leukemia is a form of blood cancer that targets white blood cells (WBC) and stands as a major cause of mortality worldwide. During the center of human bones, leukaemia is presented and provides blood cell generation with leukocytes and WBC, and if some cell comes to be blasted, then it grows a fatal illness. For that reason, the analysis of leukaemia in its initial stages aids significantly in the treatment accompanied by saving the life. At present, leukemia analysis is done by visual assessment of biomedical images of blood cells, which is time-consuming, tedious, and wants to train specialists. Consequently, the lack of an early, automatic, and effectual leukemia recognition model is a major problem in hospitals. A few automated techniques like deep learning (DL) and Machine learning (ML) methodologies for leukemia cancer identification are presented that offer remarkable and effectual results. This study develops a Robust Multimodal Fusion of Transfer Learning Framework for Leukemia Cancer Detection and Classification (RMFTLF-LCDC) algorithm. The RMFTLF-LCDC system mostly suggests identifying and classifying the existence of leukemia cancer on biomedical imaging. At first, the RMFTLF-LCDC model applies image preprocessing using a kernel correlation filter (KCF) to eliminate the noise. For the feature extraction process, the multimodal fusion of CapsNet models, including RES-CapsNet, VGG-CapsNet, and GN-CapsNet are implemented to improve the representation of features by providing more accurate initial information to subsequent capsule layers. In addition, the recurrent spiking neural network with the spiking convolutional block attention module (RSNN-CBAM) technique is performed for the leukemia cancer detection process. At last, the improved Harris hawk optimization (IHHO) approach-based hyperparameter choice can be executed to improve the classification outcomes of the RSNN-CBAM system. The efficiency of the RMFTLF-LCDC method has been validated by comprehensive studies using the benchmark image dataset. The numerical result shows that the RMFTLF-LCDC method has better performance and scalability across other recent techniques.

groups

Arwa Darwish Alzughaibi mail

link https://doi.org/10.54216/FPA.170230

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new