Serap Ercan Cömert

This study proposes an interpretable machine learning framework for predicting multiple quality outcomes in a printing production process characterized by complex and nonlinear relationships. The analysis is based on a real-world industrial dataset including key process parameters, namely Ink Weight, Top Coat Weight, Printing Speed, Varnish GC, and Temperature, along with two quality indicators: Solvent Residue and Isooctane Ratio. To capture nonlinear patterns and interaction effects, several machine learning algorithms, including K-Nearest Neighbors (KNN), Random Forest (RF), Gradient Boosting, Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Artificial Neural Network (ANN), were implemented and comparatively evaluated. The proposed framework integrates data preprocessing, polynomial feature expansion, and cross-validation-based hyperparameter optimization to ensure robustness and generalization capability. The results demonstrate that nonlinear and instance-based learning approaches outperform conventional models in representing the underlying data structure. In particular, KNN achieves the highest predictive performance across both target variables, with R² values of 0.9623 for Solvent Residue and 0.9094 for Isooctane Ratio, indicating strong agreement between predicted and observed values. To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was conducted to quantify the contribution of each input variable to model predictions. The findings reveal that process-quality relationships are highly nonlinear and governed by complex interaction effects across different operating conditions. The proposed framework contributes to the literature by enabling the simultaneous prediction of multiple quality indicators, providing a robust and interpretable decision-support tool for data-driven process optimization in printing production systems.

Recent large-scale disasters severely disrupt transportation infrastructures, making humanitarian logistics a complex decision-making problem under uncertainty and time pressure. In such environments, heterogeneous road accessibility substantially reduces the effectiveness of conventional ground-based distribution systems. Although truck–drone routing has attracted increasing attention in recent years, existing studies primarily focus on commercial delivery settings and rarely incorporate disaster-specific conditions such as disrupted accessibility and priority-sensitive demand structures. This study proposes an accessibility-aware hybrid humanitarian distribution framework integrating multiple trucks and drones for post-disaster operations. Demand points are categorized into fully accessible, partially accessible, and inaccessible regions according to post-disaster road conditions, and coordinated truck–drone operations are designed accordingly. The problem is formulated as a multi-objective optimization model that simultaneously minimizes makespan and priority-weighted total completion time. Given the NP-hard nature of the problem, a customized hierarchical Ant Colony Optimization (ACO) algorithm is developed to generate efficient routing and allocation decisions. The effectiveness of the proposed ACO framework is validated using disaster-adapted benchmark instances through comparisons with Adaptive Large Neighborhood Search (ALNS), Genetic Algorithm (GA), and Simulated Annealing (SA), demonstrating superior solution quality and robustness. The proposed framework is additionally demonstrated through a real-world case study conducted in Sakarya. Computational results indicate that the hybrid truck–drone system significantly improves operational efficiency, particularly under limited-accessibility conditions, and consistently outperforms truck-only distribution strategies. Overall, the study provides a practical, resilience-oriented, and accessibility-aware decision-support framework for enhancing the efficiency, equity, and responsiveness of post-disaster humanitarian logistics operations.

Demand, inventory, and production continuously influence one another in supply chains. Sudden demand changes, inaccurate records, production disruptions, and external factors can create irregular fluctuations. Although demand or sales data are often available, inventory and production information may be incomplete or delayed. This study applies the Extended Kalman Filter (EKF) to jointly estimate inventory, production quantity, and three model parameters using demand as the only measured variable. Its main contribution is a recursive structure that updates the state and parameter estimates with each new measurement. The method is tested on a three-variable supply chain model exhibiting chaotic behavior. Performance is assessed through numerical errors, convergence, time series, phase portraits, and bifurcation diagrams. Robustness to different initial estimates, sensitivity to process- and measurement-noise covariance settings, and performance under chaotic, semi-chaotic, and ordered regimes are also examined. Parameter errors range from 1.5% to 2.4%, while normalized errors for demand, inventory, and production remain below 3%. A larger assumed measurement-noise covariance slows convergence. Under the selected conditions, the closest agreement occurs in the chaotic regime, but this is not interpreted as a general superiority of the EKF. With real-data validation, the framework could support estimation of missing or delayed information, short-term monitoring, early warning, and decision support.

Spare parts used in renewable energy systems differ in terms of lead times, failure impacts, and production loss risks, making criticality assessment an important decision problem. Identifying critical spare parts and evaluating their implications for inventory decisions is essential for maintaining operational continuity in wind and solar energy plants. This study proposes an integrated decision-support framework for spare parts criticality assessment and inventory cost evaluation. Uncertain expert evaluations are modeled using Picture Fuzzy sets, allowing decision makers to express acceptance, rejection, and hesitation simultaneously. The relative importance of evaluation criteria is determined using the SWARA method, and spare parts are evaluated based on criteria such as price, lead time, failure frequency, and production loss cost. The Taguchi loss function is then applied to quantify the economic impact of deviations from target inventory levels. The results indicate that such deviations can lead to considerable economic losses, particularly for highly critical spare parts.

In this study, a two-stage approach based on K-Means clustering and Ant Colony Optimization (ACO) is proposed for solving the rack storage location assignment problem (SLAP) in a cable manufacturing facility where automated transport robots are used. In the first stage, 184 cables were clustered using the K-Means algorithm according to their monthly usage quantities across six production machines; the optimal number of clusters was determined as k=7 (silhouette score of 0.554) by jointly evaluating the elbow and silhouette analyses. In the second stage, a cost matrix was constructed by taking into account the usage-weighted access time of each cable to all the machines it serves, and cable-cell assignments were performed using the ACO algorithm. The ACO parameters were tuned using the Taguchi experimental design (L27 orthogonal array), and the resulting solutions were improved with Swap and Relocate local search operators. Applied to a real storage layout consisting of 18 rows × 30 columns with 504 available cells, the proposed approach reduced the total usage-weighted access cost to 1,608,326.24 usage-weighted seconds (usage count × s) after 1,000 iterations. The average usage-weighted access cost per cable was 8,740.90 usage-weighted seconds, with minimum and maximum values of 6.41 and 29,347.75 usage-weighted seconds, respectively. The results demonstrated that cables with high usage intensity were largely placed in cells closest to the machine supply points. The proposed approach, applicable to real industrial data, contributes as a data-driven decision-support tool to reducing robot movements and organizing storage areas more efficiently.

The rapid and efficient deployment of unmanned aerial vehicles (UAVs) in post-disaster scenarios is recognized as critical for effective management and logistics. The problem of selecting the most suitable UAVs for three key tasks—monitoring, light load transportation, and heavy load transportation—is addressed in this study. An extensive review of related literature was conducted to identify specific evaluation criteria tailored to each task. The UAV evaluation problem has been divided into three sub-problems, and an integrated multi-criteria decision-making (MCDM) approach is proposed within a Pythagorean Fuzzy environment for evaluation. The Step-Wise Weight Assessment Ratio Analysis (SWARA) method was employed to determine criteria weights. To rank UAVs currently available on the market, the Evaluation Based on Distance from Average Solution (EDAS) and Combinative Distance-based Assessment (CODAS) techniques were applied, with the Copeland Score used to aggregate and finalize the ranking based on each UAV’s overall performance across criteria. The results highlight key UAV choices for each task, guiding decision-makers toward UAVs that best meet the specific operational needs of post-disaster logistics. This framework provides a robust, adaptable tool for enhancing the agility and effectiveness of disaster response strategies through informed UAV selection.

COVID-19 salgınının getirdiği yaşam koşulları ile birlikte dünya ticaretine birçok olumsuz etkisine rağmen e-ticarette hızlı bir gelişmeye yol açmıştır. Talebe istinaden artan e-ticaret hacminin vazgeçilmez unsuru olan lojistik kısmı üzerine daha fazla çalışmalar ve yatırımlar yapılmaya başlanmıştır. Lojistik maliyetlerinin en büyük kısmını oluşturduğu iddia edilen son adım teslimatı üzerine yapılan çalışmalarla birlikte kısıtlı kaynaklarla yüksek miktardaki talebe cevap verirken maliyetlerin de kontrol altında tutulabilmesi için çeşitli lojistik modelleri ortaya çıktığı görülmüştür. Bu çalışmada, mobilya sektöründe e-ticaret faaliyeti yürüten bir firmanın en uygun son adım teslimatı stratejisini belirleyebilmek için çeşitli lojistik modelleri üç farklı senaryo altında incelenmiştir. Senaryoları çözümleyebilmek için p-medyan problemi, kümeleme analizi ve yerel arama destekli Clarke ve Wright tasarruf algoritmasına dayanan hiyerarşik bir optimizasyon modeli önerilmiştir. Belirlenen üç senaryodan elde edilen sonuçlar firmanın mevcut durumu ile karşılaştırılıp son adım teslimatı maliyetini minimize edecek olan en uygun seçenek yönetime sunulmuştur.

The Covid-19 pandemic has significantly impacted consumer behavior and commerce, prompting a shift towards online goods and services. The surge in demand has led to inefficiencies and disruptions, especially in the last-mile delivery (LMD) process. Because of the LMD, the final stage of the supply chain, plays a crucial role in transporting goods from businesses to consumers, challenges such as the cost inefficiencies of direct home delivery have underscored the need for innovative solutions. In this study, the collection delivery points (CDPs) approach was adopted instead of direct home delivery. It focuses on addressing these challenges by adopting service points as dynamic CDPs and handling the problem as a dynamic location routing problem (DLRP). Two solutions approaches are proposed, to select candidate depots strategically and determine efficient route configurations, to aim to minimize travel distance. One of them is a two-phased hierarchical method that starts with clustering and continues with an Ant Colony Optimization (ACO) based-hybrid algorithm, and the other one is based solely on an ACO-based hybrid algorithm. The performance of these approaches is evaluated on modified benchmark instances from the literature. It has been observed that the ACO based-hybrid algorithm is more successful in terms of total travel distance, and if an evaluation is made in terms of the number of routes, it is recommended that the results of the two-phased hierarchical method should also be considered. Furthermore, a real word case study was conducted with the proposed methods and the results were compared from different perspectives. The results corroborate the findings regarding benchmark instances, thereby providing additional validation to the results obtained.