Kocaeli University Research Information System
Not a member yet
80615 research outputs found
Sort by
Reply to Gargiulo et al.: Dose-stratified real-world effectiveness of JAK1 inhibitors in atopic dermatitis
No full textTransformer Tokenization Strategies for Network Intrusion Detection: Addressing Class Imbalance Through Architecture Optimization
No full textNetwork intrusion detection has challenges that fundamentally differ from language and vision tasks typically addressed by Transformer models. In particular, network traffic features lack inherent ordering, datasets are extremely class-imbalanced (with benign traffic often exceeding 80%), and reported accuracies in the literature vary widely (57–95%) without systematic explanation. To address these challenges, we propose a controlled experimental study that isolates and quantifies the impact of tokenization strategies on Transformer-based intrusion detection systems. Specifically, we introduce and compare three tokenization approaches—feature-wise tokenization (78 tokens) based on CICIDS2017, a sample-wise single-token baseline, and an optimized sample-wise tokenization—under identical training and evaluation protocols on a highly imbalanced intrusion detection dataset. We demonstrate that tokenization choice alone accounts for an accuracy gap of 37.43 percentage points, improving performance from 57.09% to 94.52% (100 K data). Furthermore, we show that architectural mechanisms for handling class imbalance—namely Batch Normalization and capped loss weights—yield an additional 15.05% improvement, making them approximately 21× more effective than increasing the training data by 50%. We achieve a macro-average AUC of 0.98, improve minority-class recall by 7–12%, and maintain strong discrimination even for classes with as few as four samples (AUC 0.9811). These results highlight tokenization and imbalance-aware architectural design as primary drivers of performance in Transformer-based intrusion detection and contribute practical guidance for deploying such models in modern network infrastructures, including IoT and cloud environments where extreme class imbalance is inherent. This study also presents practical implementation scheme recommending sample-wise tokenization, constrained class weighting, and Batch Normalization after embedding and classification layers to improve stability and performance in highly unstable table-based IDS problems
The Morphology of Continuity: An Analytical Documentation of Vernacular Architecture in Phoenix Rural Settlements
No full textRural architectural systems in the Mediterranean reflect a long-term entanglement between human agency, material conditions, and environmental constraints. This study uses this framework to explore architectural continuity in settlements near ancient Phoenix in Türkiye. While scholarly focus often remains on monumental ruins, it aims to examine how rural building practices, such as stone masonry, traditional carpentry, and the reuse of spolia, have persisted since antiquity. The methodology combines UAV photogrammetry, GIS analysis, and oral histories to reveal spatial patterns and craft traditions across generations. The findings show that structures are transmitted through technical knowledge, with stone and timber co-evolving with local livelihoods. By documenting the structural logic and embedded intangible knowledge of seasonal settlements like Fenaket and Büğüş, the study identifies a ‘continuity through change’ paradigm rooted in circular resilience and adaptive reuse, This study emphasizes the need for conservation strategies that integrate digital documentation with community experience to preserve the cross-border cultural landscape of the Aegean region amid environmental threats and the decline in craftsmanship, thereby sustaining it as a dynamic living culture
Investigation of broad autism phenotype levels and mind-reading from the eyes skills of parents of children diagnosed with autism spectrum disorder and language disorder
No full textObjectives: This study aimed to compare broad autism phenotype (BAP) features–considered an endophenotype of autism spectrum disorder (ASD)–and mind-reading from the eyes skills in parents of children diagnosed with ASD and language disorder (LD) with those of parents of healthy developing (HD) children, and to identify possible differences. Methods: The study included 92 children (ASD n = 31, LD n = 31, HD n = 30) and their parents (n = 184). Parents who participated completed the Autism Spectrum Quotient (ASQ) and the Reading the Mind in the Eyes Test (RMET). Results: Fathers of children with ASD and LD showed similar scores on the social skills and attention to detail subscales and scored significantly higher than fathers of HD children (p <.05). No significant differences were found in RMET scores among parent groups (mothers p =.344; fathers p =.834). Across all participants, ASQ total and social skills subscale scores were negatively correlated with RMET total scores. Conclusions: Parents of children with ASD and LD differed from controls in BAP characteristics, but RMET performance showed no group differences. These findings support previous research indicating etiological and pathological continuity between ASD and LD
A rare complication of left main bifurcation percutaneous coronary intervention: excessive longitudinal stent elongation and fracture in a hemodynamically unstable patient
No full text
A model proposal for the stardom process of televisionfigures: “The Stardom Trajectory” and representativefigures from Turkish television
No full textThis article examines how media figures who have achievedlong-term visibility on Turkish television become stars through personasthat evolve within media representations beyond fictionalroles. To explain how this star status is sustained over time, thestudy proposes a five-stage conceptual model titled the StardomTrajectory. The model draws on key theoretical approaches to stardomand celebrity developed by Dyer, Gamson, Turner, Deller, andLanger, and is further elaborated through television-specificdynamics such as repetition, cross-platform circulation, parasocialrelations, and continuity-based personality construction. The studyemploys a qualitative representational analysis grounded in StuartHall’s theory of representation and adopts a multiple-case researchdesign focusing on three figures Huysuz Virjin, Mehmet Ali Erbil,and Seda Sayan selected for their long-term presence in Turkishtelevision culture. The analysis examines their media representations,persona transformations, and screen identities across differenttelevision program formats within the framework of five stages:visibility, static figure construction, transformation, stasis, and iconization.The Stardom Trajectory model offers a multi-layered conceptualframework for understanding how television figuresachieve enduring stardom through representational continuity,familiarity, and the temporal dynamics of persona construction,contributing an original analytical perspective to studies of televisionstardom and cultural iconization.</p
Forecasting Solar Energy Production Using Artificial Neural Networks and Tyrannosaurus Optimization Algorithm
No full textAccurate forecasting of solar energy production plays a crucial role in optimizing power system reliability, scheduling, and integration of renewable energy sources into the grid. From a sustainability perspective, improved forecasting accuracy supports more efficient day-ahead planning, reduces imbalance costs, and contributes to the sustainable operation of solar energy systems. Artificial neural networks (ANNs) are widely applied for this purpose due to their capability to capture complex nonlinear relationships between meteorological variables and solar power output. However, the performance of ANNs depends on the number of layers, the number of neurons in the hidden layer, the max failure value, and the transfer function. This study proposes a hybrid forecasting model that combines artificial neural networks with the recently developed Tyrannosaurus Optimization Algorithm (TROA), a metaheuristic optimization method. The aim is to optimize the hyperparameters of artificial neural networks to minimize the Mean Absolute Percentage Error (MAPE) in solar energy forecasting. The results of the TROA were compared with other metaheuristic methods, such as Whale Optimization Algorithm (WOA), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA). The TROA gave the network structure for ANNs, which forecasted closer to the actual values than other metaheuristic methods and showed success on 105 days of the test dataset, with an MAPE rate of 3.64%. Additionally, an MAPE of 1.42% was obtained over a test period of 18 days used for out-of-evaluation, indicating competitive performance compared to the other methods. These findings highlight the effectiveness of the TROA in forecasting solar energy using ANNs
