Ahmad Yousef Abedelrahman

Ahmad Yousef Abedelrahman

Auto-generated profile, based on publicly available information

Ahmad Yousef Abedelrahman


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 December 2025 in East Journal of Computer Science
Verified
Journal Article
Open access
PDF file
AAAhmed Yousef Mohmmad Abdelrahman
Ahmed Yousef Mohmmad Abdelrahman
يعدّ التعرّف المبكر على مرض الزهايمر أمراً بالغ الأهمية من أجل إبطاء مسار التدهور المعرفي وتحسين جودة الحياة لدى المصابين بالمرض. ورغم أن صور الرنين المغناطيسي للدماغ توفّر معلومات بنيوية قيّمة قد تسهم في الكشف المبكر، إلا أن تقنيات الذكاء الاصطناعي الحديثة والواسعة الانتشار يمكن أن تعزّز هذه العملية بصورة أكبر. في هذه الدراسة، نقدّم إطاراً يعتمد على التعلّم العميق للتصنيف الموضعي وشبه المؤتمت لصور الرنين المغناطيسي إلى أربع درجات من الصحة المعرفية: عدم الإصابة بالخرف، الخرف بدرجة طفيفة جداً، الخرف بدرجة طفيفة، والخرف المعتدل. يقوم الإطار على معمارية EfficientNet-B3 وقد أظهر قدرة موثوقة على التمييز بين السمات الدقيقة المرتبطة بتطوّر التنكس العصبي في بنية الدماغ. تم تدريب النموذج على مجموعة بيانات كبيرة من صور الرنين المغناطيسي للدماغ وحقق دقة تحقق بلغت 99.4%، مع حفاظه على مستويات عالية من الدقة ومعامل F1 عبر جميع الفئات. وتشير هذه النتائج إلى أن المنهجية المقترحة قد توفّر أسلوباً موثوقاً لتحديد مستويات الصحة المعرفية، وقد تساعد الأطباء في تعزيز القرارات السريرية المبكرة. وبشكل عام، تُظهر هذه الدراسة...

30 October 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
AAAhmed Yousef Abdelrahman
Ahmed Yousef Abdelrahman
Images produced by generative models now look convincing enough that visual inspection alone is oftenunreliable. This paper examines a practical way to distinguish real photographs from AI-generated imagesusing an EfficientNet-B2 classifier supplied with complementary image information. The input includesthree RGB channels, a Fourier-frequency map, and an artifact-residual map. The evaluation uses 40,000labeled examples drawn from CommunityForensics-Small, divided into 28,000 training, 6,000 validation,and 6,000 test images. On the held-out test set, the classifier reached 94.53% accuracy (95% stratifiedbootstrap confidence interval: 93.95%–95.10%), an F1-score of 94.50% for the synthetic class, and anAUROC of 98.57%. The operating threshold was selected from validation data rather than tuned on thetest set. These results demonstrate encouraging performance within the sampled data distribution. They donot establish reliability on image generators absent from training, nor do they establish robustness to everyform of social-media processing. The study presents a reproducible detection pipeline and identifies thetests needed before the method can be considered a dependable general-purpose forensic tool.

4 November 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
AAAhmad Yousef Abedelrahman
Ahmad Yousef Abedelrahman
Phishing detectors can achieve almost perfect accuracy on benchmark datasets and still struggle with the links people encounter in practice. This paper investigates that gap and describes a hybrid framework for assessing URLs and emails through machine learning, language analysis, header inspection, and transparent rules. Our URL model combines character-level TF-IDF with 17 lexical features in a logistic regression classifier. Analysis of the 235,795 URLs in PhiUSIIL revealed pronounced differences in how the two classes were represented: every legitimate example used a simple HTTPS domain format, while 59.9% of phishing URLs included a path and 51.3% used HTTP. A classifier trained on raw URLs achieved 99.71% test accuracy, yet incorrectly marked 26 of 30 ordinary legitimate links as phishing. To limit reliance on these patterns, we trained a second classifier on 197,591 unique registered domains. On held-out domains, it achieved 86.02% accuracy, a ROC-AUC of 0.892, and 89.02% phishing precision. We then combined its predictions with URL-pattern checks, NLTK-based email cues, and SPF, DKIM, and DMARC evidence in a Flask prototype that reports a risk score and supporting reasons. The results show why dataset composition and realistic tests matter as much as headline accuracy when evaluating phishing detection.

2 November 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
AAAhmad Yousef Abedelrahman
Ahmad Yousef Abedelrahman
Reliable classification of AI-generated images requires more than high accuracy on unmodified test examples: media files may be compressed or downsampled before forensic analysis. This paper presents a controlled, paired robustness evaluation of a previously developed EfficientNet-B2 detector using RGB, fast Fourier transform (FFT), and artifact-oriented image features. The existing checkpoint, validation-selected classification threshold, and fixed set of 6,000 held-out images were reused without training or test-time threshold adjustment. Six transformed conditions were evaluated against original inputs: JPEG qualities 90, 70, and 50, and bicubic downsampling to maximum dimensions of 256, 128, and 64 pixels followed by reconstruction to the original dimensions. The baseline achieved 94.53% accuracy and 0.9857 area under the receiver operating characteristic curve (AUROC). JPEG variants produced accuracies between 94.38% and 94.63%, with paired 95% bootstrap intervals of accuracy change spanning zero. By contrast, downsampling to 256, 128, and 64 pixels yielded accuracies of 93.15%, 91.73%, and 74.40%, respectively. At 64 pixels, the accuracy difference was −20.13 percentage points (95% paired bootstrap interval approximately −21.12 to −19.12), and AUROC fell to 0.8967. The evidence indicates substantially greater sensitivity to severe spatial resolution loss than to the tested JPEG compression settings. Because the existing training procedure incorporates JPEG augmentation, these findings characterize this detector and evaluation protocol rather than establish universal robustness across detectors or unseen transformations.

25 February 2026 in Midocean Journal for Research and Studies
Verified
Journal Article
Closed access
PDF file
AAAhmed Yousef Abdelrahman
Ahmed Yousef Abdelrahman
Objectives: This research aims to identify subtle signs in Arabic Sign Language to overcome the communication barrier for the deaf and hard-of-hearing community. This study focuses on Arabic-speaking communities, particularly in Saudi Arabia, and explores dialect differences in Sign Language, appropriateness, and generalization. Methodology : A convergent mixed-methods design is used, combining quantitative evaluation of deep learning models with assessment by deaf and hard-of-hearing individuals. This design also demonstrates the relevance of having an available system to translate Arabic Sign Language. Results : The results show a strong correlation between gesture recognition accuracy and visual generalization, with a 98% improvement in recognizing dynamic gestures. Conclusion : The paper outlines directions for future research to enhance assistive communication systems, focusing on movement and body language analysis. The main considerations include precision mechanisms, potential applications, and the possibility of real-time adaptation. The paper also suggests that the software can be developed for mobile or desktop platforms to facilitate communication for the deaf in the Arabic-speaking world.