Çiğdem Acı Çiğdem Acı MÜHENDİSLİK FAKÜLTESİ BİLGİSAYAR MÜHENDİSLİĞİ BÖLÜMÜ BİLGİSAYAR YAZILIMI ANABİLİM DALI
Araştırmacının yayınlarını daha detaylı bir şekilde aramak veya filtrelemek için buraya tıklayabilirsiniz.

Morphological and structural complexity analysis of low-resource English-Turkish language pair using neural machine translation models

Acı, Mehmet | Vuran Sarı, Nisa | Acı, Çiğdem

Neural machine translation (NMT) has achieved remarkable success in high-resource language pairs; however, its effectiveness for morphologically rich and low-resource languages like Turkish remains underexplored. As a highly agglutinative and morphologically complex language with limited high-quality parallel data, Turkish serves as a representative case for evaluating NMT systems on low-resource and linguistically challenging settings. Its structural divergence from English makes it a critical testbed for assessing tokenization strategies, attention mechanisms, and model generalizability in neural translation. This study investigates the comparative performance of two prominent NMT paradigms—the Transformer architecture, and recurrent-based sequence-to-sequence (Seq2Seq) models with attention for both English-to-Turkish and Turkish-to-English translation. The models are evaluated under various configurations, including different tokenization strategies (Byte Pair Encoding (BPE) vs. Word Tokenization), attention mechanisms (Bahdanau and an exploratory hybrid mechanism combining Bahdanau and Scaled Dot-Product attention), and architectural depths (layer count and attention head number). Extensive experiments using automatic metrics such as BiLingual Evaluation Understudy (BLEU), M...

A Feature Selection-Based Multi-Stage Methodology for Improving Driver Injury Severity Prediction on Imbalanced Crash Data

Çiğdem İnan ACI | Gizen Mutlu | Murat Ozen | Esra Sarac | Vahide Nida Kılıç Uzel

Predicting driver injury severity is critical for enhancing road safety, but it is complicated because fatal accidents inherently create class imbalance within datasets. This study conducts a comparative analysis of machine-learning (ML) and deep-learning (DL) models for multi-class driver injury severity prediction using a comprehensive dataset of 107,195 traffic accidents from the Adana, Mersin, and Antalya provinces in Turkey (2018–2023). To address the significant imbalance between fatal, injury, and non-injury classes, the hybrid SMOTE-ENN algorithm was employed for data balancing. Subsequently, feature selection techniques, including Relief-F, Extra Trees, and Recursive Feature Elimination (RFE), were utilized to identify the most influential predictors. Various ML models (K-Nearest Neighbors (KNN), XGBoost, Random Forest) and DL architectures (Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN)) were developed and rigorously evaluated. The findings demonstrate that traditional ML models, particularly KNN (0.95 accuracy, 0.95 F1-macro) and XGBoost (0.92 accuracy, 0.92 F1-macro), significantly outperformed DL models. The SMOTE-ENN technique proved effective in managing class imbalance, and RFE identified a critical 25-feature sub...

Enhanced Multi-Class Driver Injury Severity Prediction Using a Hybrid Deep Learning and Random Forest Approach

Çiğdem İnan Acı | Gizen Mutlu | Murat Ozen | Mehmet Acı

Predicting driver injury severity and identifying factors influencing crash outcomes are crucial for developing effective traffic safety measures. This study focuses on estimating driver injury severity (uninjured, injured, or killed) and determining critical factors affecting crash outcomes. A hybrid framework combining Deep Neural Networks (DNNs) and Random Forest (RF) is proposed, where a DNN extracts features and RF performs the final classification, leveraging ensemble methods. The results were compared with those of well-known methods (e.g., kNN, XGBoost), with the hybrid approach achieving the best performance (0.92 accuracy, 0.89 F1-macro, 0.91 F1-micro scores) in predicting injury severity. The results showed that crash type, vehicle type, driver fault, intersection type, season, time, and road type had the greatest impact, while factors like pavement condition and driver gender had minimal influence. To the best of our knowledge, this is the first study to combine DNN-based feature extraction with RF classification in the context of traffic injury severity prediction. The framework offers a new approach for drivers and policymakers, providing a deeper understanding of driver injury severity prediction and its underlying factors.

Windows Malware Detection via Enhanced Graph Representations with Node2Vec and Graph Attention Network

Nisa Vuran Sarı | Mehmet Acı | Çiğdem İnan Acı

As malware has become increasingly complex, advanced techniques have emerged to improve traditional detection systems. The increasing complexity of malware poses significant challenges in cybersecurity due to the inability of existing methods to understand detailed and contextual relationships in modern software behavior. Therefore, developing innovative detection frameworks that can effectively analyze and interpret these complex patterns has become critical. This work presents a novel framework integrating API call sequences and DLL information into a unified, graph-based representation to analyze malware behavior comprehensively. The proposed model generates initial embeddings using Node2Vec, which uses a random walk approach to understand structural relationships between nodes. Graph Attention Network (GAT) then enhances these initial embeddings, which utilizes attention mechanisms to incorporate contextual dependencies and enhance semantic representations. Finally, the enhanced embeddings are classified using Convolutional Neural Network (CNN) and Gated Recurrent Units (GRU)s, a custom hybrid CNN-GRU-3 deep learning-based model capable of effectively modeling sequential patterns. The dual role of GAT as a classifier and feature extractor is also analyzed to evaluate its impa...

Deep Learning-Based Prediction Models for the Detection of Vitamin D Deficiency and 25-Hydroxyvitamin D Levels Using Complete Blood Count Tests

Acı, Çiğdem | Acı, Mehmet

Vitamin D (VitD) is an essential nutrient that is critical for the well-being of both adults and children, and its deficiency is recognized as a precursor to several diseases. In previous studies, researchers have approached the problem of detecting vitamin D deficiency (VDD) as a single ”sufficient/deficient” classification problem using machine learning or statistics-based methods. The main objective of this paper is to predict a patient’s VitD status (i.e., sufficiency, insufficiency, or deficiency), severity of VDD (i.e., mild, moderate, or severe), and 25-hydroxyvitamin D (25(OH)D) level in a separate deep learning (DL)-based models. An original dataset consisting of complete blood count (CBC) tests from 907 patients, including 25(OH)D concentrations, collected from a public health laboratory was used for this purpose. CNN, RNN, LSTM, GRU and Auto-encoder algorithms were used to develop DL-based models. The top 25 features in the CBC tests were carefully selected by implementing the Extra Trees Classifier and Multi-task LASSO feature selection algorithms. The performance of the models was evaluated using metrics such as accuracy, F1-score, mean absolute error, root mean square error and R-squared. Remarkably, all three models showed satisfactory results when compared to the ...