2026-03-15
Hayder N. Jasim, Wesam M. Jasim, Mohammed S. Ibrahim
Breast cancer is a major global health concern, highlighting the need for accurate and efficient diagnostic solutions rather than persistent issues with detection accuracy. This study presents an enhanced machine learning framework to improve breast cancer classification by addressing key limitations: Class imbalance, irrelevant features, and suboptimal hyperparameters. Adaptive synthetic sampling (ADASYN) was used to balance class distribution and various feature selection techniques. Univariate Selection and recursive feature elimination improved feature relevance, and arctic puffin optimization (APO) was applied for hyperparameter tuning. Multiple classifiers were evaluated using the Wisconsin Diagnostic Breast Cancer dataset. The random forest (RF) with ADASYN approach, optimized using APO, achieved outstanding results – 99.53% accuracy, 100% precision, 99.07% recall, and 99.53% F1-score – with only one misclassification out of 569 samples. This framework, while not modifying ADASYN or RF algorithms themselves, significantly enhances diagnostic performance and serves as a robust foundation for clinical decision support systems.
2026-02-11
Soran A. Hamad, Kayhan Z. Ghafoor
Distributed denial of service (DDoS) attacks are a significant danger to network security, with SYN flood assaults being particularly known for exploiting the transmission control protocol (TCP) handshake to deplete server resources. This review paper analyzes the current research on classifying DDoS attacks using machine learning (ML) approaches, with a focus on SYN f lood scenarios. Traditional algorithms such as XGBoost, Random Forest, and k-Nearest Neighbors are examined alongside modern deep learning methods such as convolutional neural networks and long short-term memory networks. Deep learning, noted for its capacity to automatically learn complex properties from data, is particularly effective in dynamic contexts like the internet of things. The review analyzes the usefulness of various strategies, obstacles in feature engineering and model training, and their implications for real-time detection. This study presents a comprehensive overview of the accomplishments in employing ML and deep learning for TCP SYN flood attack classification and exposes gaps in the field that indicate options for further research.
2026-01-30
Sheelan A. Ahmed, Hirsh M. Majid
Modifying the asphalt binder is one technique for improving asphalt pavement performance. When nanoparticles are utilized to modify asphalt binders, they exhibit unique and significant properties. Without a doubt, silica is the most commonly added element as nanoparticles to asphalts. The main objective of this study was to investigate the effect of mixing speed used to produce nanosilica (NS)-modified-asphalt cement on conventional physical properties of asphalt cement and prepare a model to predict the physical properties of NS-modified asphalt cement. For this purpose, four different nanosilica contents (1.5%, 3%, 4.5%, and 6% of the asphalt cement weight) were utilized. Each content was mixed at a temperature of 160°C using a high shear mixer set to four different speeds (1,000 rpm, 2,000 rpm, 3,000 rpm, and 4,000 rpm) for 30 min. Conventional asphalt cement tests, including penetration, softening point, rotational viscosity, and ductility tests, are conducted to evaluate the nanosilica-modified asphalt cement properties. The results showed that higher NS contents improve the physical properties of modified asphalt cement, but the improvement is more noticeable when mixing at a faster speed, as this lowers penetration and rotational viscosity and raises the softening point. By utilizing the Minitab statistical program, the physical properties prediction models of the NSmodified asphalt showed a direct correlation between input data (NS content, NS average size, mixing temperature, mixing rotational speed, mixing time, and neat asphalt properties) and predictive models, demonstrated by a high R2 value and acceptable standard deviation values.
2026-01-29
Soleen J. Ibrahim, Shaheen A. Abdulkareem, Ahmad B. Al-Khalil
A contactless card is an easy and straightforward way to make a purchase that takes only a few seconds without requiring a Personal Identification Number (PIN) or signature. However, not requiring a PIN or signature on a contactless bank card makes it vulnerable to fraud attacks when losing or stealing the card. This article delivers a new model which is developed for securing a bank contactless card with fingerprint authentication. The suggested model is the creation of a new fingerprinting algorithm that combines with the virtual contactless card. The fingerprint recognition algorithm employs image processing methods to enhance and extract features in order to compare fingerprint impression images. The performance of the proposed model is evaluated based on two metrics: false acceptance rate and false rejection rate. There are five scenarios to test and evaluate the proposed model. The findings establish that the developed system enhances the process of embedding biometrics (fingerprints) to non-contact smart card and a user-friendly experience.
2026-01-20
Ali H. Ibrahem, Adnan M. Abdelazeez
In response to the demographic bias commonly observed in facial beauty prediction (FBP) models due to the underrepresentation of certain ethnic groups, KurdFace-1000 is introduced as a novel and balanced dataset developed as a field of research community. This dataset comprises 1000 color facial images of individuals from the Kurdistan Region of Iraq, evenly distributed across gender (500 male, 500 female) and facial expression (500 smiling, 500 non-smiling), and includes both frontal and profile views. Within the results of males, 200 are smiled and 300 are non-smiled and for the results of females, 300 are smiled and 200 are non-smiled. Moreover, each image is annotated with three key attributes: A beauty score on a [1–5] scale rated by five independent human raters, binary gender, and smile expression. KurdFace-1000 is the first dataset specifically designed to represent Kurdish facial features in FBP tasks, aiming to reduce ethnic bias and improve model performance for underrepresented populations. With a balanced structure and diverse annotations, the dataset supports various computational paradigms, including classification and regression, and serves as a critical step toward building culturally aware and inclusive deep learning models in FBP.
2025-02-05
Sardar K. Jabrw, Qusay I. Sarhan
Large language models (LLMs), also known as generative AI, have transformed code generation by translating natural language prompts into executable code. Yet, their capabilities in generating code for resource-constrained devices such as Arduino, which are used in the Internet of Things and embedded systems, remained underexplored. This study evaluates six state-of-the-art LLMs for generating correct, efficient, and high-quality Arduino code. The evaluation was performed across five dimensions, namely functional correctness, runtime efficiency, memory usage, code quality, similarity to human-written code, and multi-round error correction. The results reveal that ChatGPT-4o achieves the highest zero-shot functional correctness and aligns closely with human code in readability and similarity. On the other hand, Gemini 2.0 Flash generates faster-executing code but at the cost of higher code complexity and lower similarity. DeepSeek-V3 balances correctness with superior flash memory optimization, whereas Claude 3.5 Sonnet struggles with prompt adherence. Finally, multi-round error correction improves correctness across all six models. Overall, the f indings underscore that none of the evaluated LLMs consistently outperforms all evaluation criteria. Hence, model choice must align with project priorities; as shown, ChatGPT-4o excels in functional correctness, whereas Gemini 2.0 excels in execution time, and DeepSeek-V3 in memory efficiency. This study provides a systematic evaluation of code generated with LLMs for Arduino, which, to the best of our knowledge, has not been previously studied across multiple models and performance metrics, thereby establishing a foundation for future research and contributing to enhancing the trustworthiness and effectiveness of LLM-generated code.