DenseNet201-Based Deep Transfer Learning Framework for Brain Tumor Classification in MRI Scans

Doaa Sami Khafaga

doi:https://doi.org/10.54216/JAIM.100204

Journal of Artificial Intelligence and Metaheuristics | 2025

Network Requests Classification Using Advanced Metaheuristic Optimization for Enhanced Network Security Systems

The importance of network security has greatly increased in the continuously changing modern digital environment. Network security is a multi-layered defense mechanism that protects networks, data, and systems from unauthorized access and other malicious activities. As cyber threats become more advanced, traditional protective measures may no longer be adequate, and intrusion-detection systems must rapidly evolve. This research proposes BBERSC (Balance Between Al-Biruni Earth Radius Optimization and Sine Cosine Algorithm), which combines the local-search capability of Al-Biruni Earth Radius Optimization with the probabilistic improvement of the Sine Cosine Algorithm. The approach tunes Random Forest hyperparameters, including the number of trees, maximum depth, and minimum samples, to improve pattern recognition in network traffic. ANOVA and Wilcoxon signed-rank tests validate the effectiveness of the models and algorithms and assess differences between them. The Random Forest model within the BBERSC framework achieves an accuracy of 0.9901719, exceeding the other evaluated machine-learning algorithms.

groups

Marwa M. Eid mail

link https://doi.org/10.54216/JAIM.100104

Volume & Issue

Vol. Volume 10 / Iss. Issue 1

Details open_in_new

Journal of Artificial Intelligence and Metaheuristics | 2025

Somersaulting Spider Optimizer (SSO): A Nature-Inspired Metaheuristic Algorithm for Engineering Optimization Problems

The growing complexity of engineering optimization problems has revealed significant limitations in traditional mathematical programming approaches, necessitating the development of innovative metaheuristic algorithms capable of handling high-dimensional, multi-modal, and discontinuous objective functions. This paper presents the Somersaulting Spider Optimizer (SSO), a novel bio-inspired metaheuristic algorithm that draws inspiration from the extraordinary locomotion mechanisms of the somersaulting spider, a desert-dwelling arachnid species renowned for its acrobatic somersaulting capabilities. The proposed algorithm incorporates dual movement mechanisms that effectively balance global exploration through somersaulting behavior and local exploitation via controlled rolling movements. A distinctive feature of SSO is its adaptive energy-management system, which dynamically regulates exploration–exploitation transitions based on solution-improvement patterns and stagnation detection. Complementary adaptive parameters balance global search and local refinement throughout optimization. The algorithm was evaluated on four engineering design benchmarks: pressure vessel, welded beam, three-bar truss, and cantilever beam. Comparisons with the Genetic Algorithm (GA), Whale Optimization Algorithm (WOA), Harris Hawks Optimization (HHO), and Bat Algorithm (BAT) report that SSO outperforms these methods on the test problems. ANOVA and Wilcoxon signed-rank tests assess the significance of the reported performance differences. The results identify SSO as an effective optimization method, with improvements in solution quality, convergence stability, and computational efficiency.

groups

Ahmed Mohamed Zaki mail -

Hala B. Nafea mail -

Hossam El-Din Moustafa mail -

El-Sayed M. El-Kenawy mail

link https://doi.org/10.54216/JAIM.100105

Volume & Issue

Vol. Volume 10 / Iss. Issue 1

Details open_in_new

Journal of Artificial Intelligence and Metaheuristics | 2025

Hybrid CNN–LSTM Architecture for OCT Retinal Disease Classification

Optical coherence tomography (OCT) provides detailed retinal images, but their interpretation can be difficult across clinical settings and patient groups. This paper presents a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture for classifying retinal OCT images into eight disease categories. Two parallel convolutional branches extract spatial features; their outputs are pooled, reshaped into a sequence, and processed by an LSTM layer to model inter-feature relationships before dense classification. The model is evaluated on 24,000 images and compared with ResNet50, MobileNetV2, InceptionV3, and Xception. The proposed architecture achieves 93.00% accuracy, 94% precision, 93% recall, and a 0.93 F1-score, outperforming the compared models. The results indicate that combining convolutional feature extraction with sequence modeling is a promising approach to automated retinal disease classification.

groups

Ehsan Khodadadi mail -

P. K. Dutta mail -

Amel Ali Alhussan mail -

Marawa Metwally mail

link https://doi.org/10.54216/JAIM.100201

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2026

EEG Signal Classification for Mental States Using Deep Learning

In recent years, EEG based recognition and characterization of brain states has received much interest due to the advances in deep learning and machine learning methods. The non-invasive and highly inexpensive activity of EEG presents a patient with details concerning the activity and the conditions of the brain. The synthesis of artificial intelligence (AI) models (convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and collaborative knowledge options has been explored in a series of studies that recognize the mental state accurately in a large number of cases. The literature focuses on introducing strong, explainable models as well as on multimodal data to boost classification accurateness and reliability. The results are a 1D CNN and a LSTM network were trained separately and in a hybrid, architecture (CNN-LSTM) to classify the EEG signals. The models were appraised using accurateness, accuracy, recollection, F1-score, and confusion matrix analysis.

groups

Abdulrahman W. H. Al-Askari mail

link https://doi.org/10.54216/FPA.210220

Volume & Issue

Vol. Volume 21 / Iss. Issue 2

Details open_in_new

Journal of Intelligent Systems and Internet of Things | 2025

Multi-View Feature Learning Approach in Deep Learning Model for Improving Endometrial Cancer Detection from Medical Images

An accurate diagnosis of Endometrial Cancer (EC) is crucial for gynecologists, as different types may require specific treatments. Radiomics, a quantitative method, can help analyze and quantify image heterogeneity, aiding in lesion diagnosis. Previous research introduced a Transformer-based Semantic-Aware U-Net with Deep Endometrial Cancer Prediction (TSA-UNet-DeepECP) to segment and classify EC stages in Magnetic Resonance Imaging MRI scans. However, the heterogeneous properties of input scans can affect the DeepECP model's performance. Hence, this study presents the TSA-UNet with an Improved DeepECP model (TSA-UNet-IDeepECP) for EC stage classification. This IDeepECP model incorporates a multi-view learning approach, combining local 2D MRI image information with global 3D MRI image information. First, the endometrium MRI scans are collected, augmented, and segmented using the TSA-UNet model. Various Deep Learning (DL) models, one for 2D and one for 3D, are fed the segmented images. In contrast to the 3D view model, which collects global information from 3D MRI images, the 2D view model primarily recovers local features from 2D MRI data. The multi-view DeepECP model is trained using these combined characteristics. A Fully Connected (FC) layer and the softmax classifier are used for classifying EC stages using the combined features. When compared to traditional models, a TSA-UNet-IDeepECP model achieves better performance in EC detection from MRI images.

groups

Karthick Natarajan mail -

Nithya Palanisamy mail

link https://doi.org/10.54216/JISIoT.170215

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new

Fusion: Practice and Applications | 2026

Identification of Post Flood Water Level Severity through UAV Images Using Attention Based Deep Learning Techniques

Floods are among the most devastating natural disasters, causing widespread damage to infrastructure, homes, and human lives. Rapid assessment of flood severity is critical for effective disaster response and resource allocation. This study explores several deep learning approaches for flood water level classification using UAV imagery. A curated dataset of 2,000 UAV images from diverse regions, including India, the United States, and Brazil, was developed and augmented to improve generalization. Multiple architectures were evaluated, including pre-trained CNNs, ResNet50v2, MobileNetv2, Vision Transformers, and Swin Transformers, with and without the Convolutional Block Attention Module (CBAM) and adaptive learning strategies. Experimental results reveal that integrating Vision Transformers with CBAM achieves the highest classification accuracy of 90.6%, while a hybrid CNN–Vision Transformer model further improves performance to 92.3%. These findings highlight the potential of attention-based hybrid models for precise flood severity mapping. The proposed framework can aid rescue teams and disaster management authorities by prioritizing high-risk areas, enabling faster response and optimized allocation of resources during emergency operations.

groups

Sanket S Kulkarni mail -

Ansuman Mahapatra mail

link https://doi.org/10.54216/FPA.210221

Volume & Issue

Vol. Volume 21 / Iss. Issue 2

Details open_in_new

Journal of Cybersecurity and Information Management | 2026

Empirical Analysis of Computationally Intelligent Technique for Software Risk Prediction

Software development is inherently associated with a high degree of uncertainty, often arising from unforeseen activities during different phases of the SDLC. As software systems expand in scale and complexity, the likelihood of failures and project delays also increases. Such situations, which are usually not anticipated, are known as software risks. They arise due to different reasons, which affect activities like essentials of engineering, making, putting into usage, and test. These risks need to be identified and managed in the initial phase for delivering software-related products that are both excellent and can be relied upon. While it has been standard practice in assessing software risks to depend upon human skills and previous experiences, it has been observed they lead to issues in consistency and often are reported to be unreliable. The current study is an attempt to tackle this issue through usage of predictive models that have their roots in machine learning (ML). Borrowing from existing data, software risks are identified and classified through five popular machine-learning tools. To improve correctness and make it more robust, selection techniques of selection with multiple features are implemented. Among the other models, the Support Vector Machine (SVM) exhibited the maximum performance, achieving a classification accuracy of approximately 80%, with a precision of 84%, recall of 80%, and an F1 score of 80%. In terms of performance, Mutual Information was found to be best in methods of applied feature selection. The study indicates the ability of ML based methods in predicting and managing software risks. Additionally, this research highlights the potential of computationally intelligent techniques to assist project managers in early risk identification, proactive decision-making and enhancing the overall success rate of s/w projects.

groups

Mohd Shabbir mail -

Rakesh Kumar Yadav mail -

Mohd Waris Khan mail -

Hitendra Singh mail

link https://doi.org/10.54216/JCIM.170214

Volume & Issue

Vol. Volume 17 / Iss. Issue 2

Details open_in_new

Journal of Artificial Intelligence and Metaheuristics | 2025

Hierarchical Clustering of Global COVID-19 Statistics: Comparative Insights from Pandemic Indicators

Hierarchical clustering is applied to global COVID-19 data from the Worldometer tracker, whose final update was April 13, 2024, to partition countries into clusters based on epidemiological indicators. Total cases, deaths, recoveries, active cases, tests, population, and per-million measures are standardized and analyzed using agglomerative hierar-chical clustering with Ward linkage. The evaluation yields an average Silhouette score of 38.5%, a Davies–Bouldin value of 0.87, and a Calinski–Harabasz value of 77.6. Dendrograms and principal component analysis (PCA) projections identify four clusters, reflecting differences in pandemic impact and response. The analysis finds that high-burden clusters account for almost 45% of global deaths, while low-burden clusters predominate in more than 40% of nations with fewer than 100,000 accumulated cases. The results illustrate how hierarchical clustering can support comparative public-health analysis by providing quantitative estimates of epidemiological patterns across countries.

groups

Noor Razzaq Abbas mail -

Ghassan AL-Thabhawee mail -

Isam Bahaa Aldallal mail -

Mostafa Abotaleb mail -

Klodian Dhoska mail

link https://doi.org/10.54216/JAIM.100202

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new

Journal of Artificial Intelligence and Metaheuristics | 2025

Optimizing CO2 Emission Forecasting from Vehicles Using Deep Learning and the Football Optimization Algorithm

Accurate prediction of CO2 emissions from vehicles is essential for environmental regulation and sustainable transport design. Existing models often suffer from limited accuracy because of suboptimal hyperparameter configurations. This study enhances CO2 emission forecasting by combining deep learning with metaheuristic optimization. An attention-based Encoder LSTM (EALSTM) model is trained on Canadian vehicle-emissions data, with hyperparameters tuned using the Football Optimization Algorithm (FbOA), inspired by cooperative team dynamics in football. Comparative evaluation against eight other optimizers shows that FbOA achieves the best performance. The optimized FbOA–EALSTM model yields an RMSE of 0.00349, MAE of 0.00010, and R2 of 0.984, outperforming the alternatives reported in the study. These results demonstrate the potential of domain-inspired metaheuristics to improve prediction accuracy. The proposed framework offers a scalable solution for emissions modeling and supports data-driven environmental policy and intelligent vehicle technologies.

groups

Omnia M. Osama mail -

El-Sayed M. El-Rabaie mail -

Marwa M. Eid mail

link https://doi.org/10.54216/JAIM.100203

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new

Journal of Artificial Intelligence and Metaheuristics | 2025

DenseNet201-Based Deep Transfer Learning Framework for Brain Tumor Classification in MRI Scans

The classification of brain tumors is crucial in the context of early intervention, as appropriate and timely diagnosis can significantly influence treatment plans and patient outcomes. Radiologists have long relied on their own judgment to read medical images, a process that can be subjective and time-consuming and can lead to inter-observer variability. Artificial intelligence (AI), and particularly deep learning (DL), has transformed medical imaging and could automate diagnosis by providing prompt, reliable, and unbiased assessments. Despite this progress, many AI-based systems remain constrained in their classification accuracy and robustness across datasets. This paper presents a deep-learning model designed to improve brain-tumor classification. Its preprocessing and normalization steps reduce noise and equalize image intensity to support feature extraction. The proposed method adapts DenseNet201 for four-class classification and replaces its conventional output layer with Global Average Pooling (GAP), aiming to address vanishing gradients and overfitting while improving feature aggregation and generalization. The model was evaluated on a public Figshare dataset and a local dataset comprising 3,504 T1-weighted contrast-enhanced MRI images. It achieved 100% overall accuracy, outperforming the comparison methods. These results suggest that architectural modifications combined with preprocessing can support automated brain-tumor classification under the evaluated conditions.

groups

Doaa Sami Khafaga mail

link https://doi.org/10.54216/JAIM.100204

Volume & Issue

Vol. Volume 10 / Iss. Issue 2

Details open_in_new