IJEEC - INTERNATIONAL JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTING

Application of DCT Compression on Images in IoT Systems

Nevena Jeftenić — 2026-09-14

Through the integration of sensors, software, and wireless connectivity into everyday activities, the Internet of Things (IoT)
creates intelligent ecosystems whose applications are inevitably permeating all segments of life, including smart cities, agriculture, healthcare, and manufacturing. Such systems frequently utilize image and video data as information sources for machine learning and analysis, which are essential for anomaly detection, security surveillance, facial recognition, traffic optimization, and—in the case of this study—monitoring bee activity within a beehive. Given that digital images contain vast amounts of data transmitted through digital processing systems, it is vital to efficiently address the challenges of data transmission within resource-constrained IoT networks. The primary objective of the research presented in this paper is to demonstrate that the application of DCT (Discrete Cosine Transform) compression on images can significantly reduce data volume while simultaneously preserving the image quality required for further application. To evaluate the compression, DCT was applied to JPEG images obtained from smart beekeeping system cameras, utilizing the Python programming language with OpenCV and NumPy libraries. The images were sourced from publicly available datasets. The analysis involved applying DCT compression across six defined quality levels: 1, 5, 20, 50, 80, and 100, covering a range that allows for the assessment of the trade-off between compression ratio and image quality. As expected, the lowest quality factor yields the smallest file sizes but results in the lowest image quality. However, it is significant that even with a moderate factor of 20, substantial optimization is achieved — a fivefold reduction in data while maintaining satisfactory image quality for further processing. The optimal result is achieved at a factor of 50, which provides a twofold reduction in data and an output image where quality differences, compared to the original, are indiscernible to the naked eye.

End-to-End Latency Decomposition in AI-Driven Web Applications: Rethinking Infrastructure in LLM Based Systems

Jasna Hamzabegović — 2026-09-14

The increasing integration of artificial intelligence into web applications, particularly through large language models (LLMs),
has fundamentally reshaped the performance characteristics of modern systems. Unlike traditional architectures, where latency is
primarily determined by backend infrastructure, AI-driven applications operate as multi-stage pipelines involving orchestration logic, network communication, and external model inference. This paper introduces an end-to-end latency decomposition framework for analyzing performance in AI-powered web applications. A controlled experimental study is conducted using two production-equivalent implementations deployed in serverless and virtual private server (VPS) environments. The methodology distinguishes between full stack execution, including LLM inference, and infrastructure-only scenarios, enabling precise isolation of latency contributions across infrastructure, application, and model layers. The results indicate that in full-stack scenarios, model-related latency dominates system performance, accounting for approximately 85% of total response time, thereby minimizing the impact of infrastructure differences. In contrast, infrastructure-only scenarios reveal significant performance variations between deployment environments. These findings challenge infrastructure-centric optimization approaches and demonstrate the need for system-level performance evaluation in LLM based applications. The proposed framework provides a practical methodology for identifying performance bottlenecks and offers actionable insights for optimizing AI-driven web systems.

Enhancing MLOps for Educational Data Mining: A Comparative Study of Weka and KNIME in Model Lifecycle Management

Ognjen P. Tomić — 2026-09-14

This paper presents a comparative case study of Weka and KNIME for supporting Machine Learning Operations principles in Educational Data Mining. The goal of the study is to evaluate how two widely used data mining platforms support both
analytical modeling and model lifecycle management in an educational data analysis context. The study uses the
Ocisceni_Demografski_Podaci.csv dataset, containing approximately 15,000 demographic and education-related records from 42
municipalities. For the experimental evaluation, a cleaned analytical subset of 256 instances and 12 attributes was prepared and used consistently in both platforms. Both platforms were evaluated using the same preprocessing logic, classification, clustering,
visualization, and lifecycle-management tasks. The analytical evaluation included Naive Bayes, k-nearest neighbors, K-Means, and
FarthestFirst algorithms, while the lifecycle evaluation assessed workflow reproducibility, pipeline automation, deployment readiness, and integration extensibility using a five-level maturity rubric. Weka classifier performance was evaluated using stratified 10-fold cross-validation, whereas KNIME classifier results were obtained through a hold-out workflow and are interpreted within the corresponding platform-specific evaluation setting. The results show that Weka provides strong support for rapid algorithmic experimentation and model comparison, with the evaluated classifiers achieving classification accuracy between 80.08% and 82.81%, with KNN (k = 10) producing the highest accuracy under the selected evaluation protocol. In contrast, KNIME provides stronger support for reproducible workflow construction, automation, integration with external systems, and operationalization through server-or hub-based infrastructure. The findings indicate that neither platform is universally superior; rather, their usefulness depends on the phase of the machine learning lifecycle. The study concludes that educational institutions can benefit from a hybrid approach in which Weka is used for early-stage model exploration, while KNIME is used for workflow reconstruction, automation, reporting, and reproducible operationalization.

Product Structure: A Data Model in the Context of Small and Medium-Sized Enterprises (SMEs)

Miroslav Dragić — 2026-09-14

In contemporary business environments—particularly amid accelerated digitalization—the capability for systematic
management of information on products and production processes has become pivotal to achieving competitive advantage. For
manufacturing small and medium-sized enterprises (SMEs), which operate under constraints in informational, material, and financial resources and under constant pressure to reduce costs and shorten delivery lead times, a clearly structured approach to product data management constitutes a key lever for achieving efficiency, quality, and market flexibility. In this context, the product structure component (PSC) and the corresponding data model are prerequisites for consistent, scalable, and interoperable information management throughout the entire product life cycle (from concept, development, and production to maintenance and recycling). Within this work, different approaches to defining PSC and the associated data models are considered, with a particular focus on how SMEs can achieve consistent management of product structure data, change and revision control, support for variant management, along with a rational selection of database technology and system architecture. This paper examines different approaches to defining product-structure data, including relational, object-oriented, and hybrid (multi-model) models. Different approaches to product data modeling for a manufacturing company are presented. The data model will include product structure data (engineering BOM), customer orders, and document management models. Finally, a brief overview of the advantages and disadvantages of each of the proposed solutions is given.

A new solution for the control system of the cross cutter in the board factory

Neša Rašić — 2026-09-14

The paper presents a new control system implemented in the phase of reconstruction and modernization of the Valmet 14.5
transverse cutter in the cardboard factory "Umka" near Belgrade, Serbia [1]. The new control system was implemented using a PLC
with integrated PROFInet and PROFIBUS communication protocols. Transverse cutters are used for transverse cutting of cardboard or paper tape into pieces of predetermined length, when unrolling it from the roll. The basic requirements that must be met are: reliable operation, work with high tape speeds in the cutting process, high accuracy of the length of the cut pieces, quality cutting, without creasing, tearing, or crushing at the cutting site. In the operation of transverse cutters, the strip being cut reaches the flying shears at the desired speed, which is determined by the speed of rotation of a pair of rollers called the press (press drive). Flying scissors consist of two blades placed perpendicular to the movement of the tape, the length of blades corresponds to the maximum width of the tape being cut (shear drive). The paper also presents the characteristics and technical requirements of the basic functional units of the transverse cardboard tape cutter from the point of view of control. The synchronized operation [2, 3] of the main electric drives powered by frequency converters is described and analyzed, which ensures the adjustment of the peripheral speed of the cutter blades, which is necessary for proper operation. The paper presents the results recorded at the reconstructed transverse cutter plant.

Edge-Aware Graph Neural Network Baselines for Protein Function Prediction on OGBN-Proteins

Aleksandar Stanković — 2026-09-14

This paper presents an engineering-oriented study of edge-aware graph neural network baselines for protein function
prediction on the OGBN-Proteins benchmark. The benchmark represents a large protein-protein interaction graph whose edges carry eight-dimensional association evidence and whose proteins have 112 functional labels. The study focuses on practical design choices that strongly affect reproducibility and deployment cost: construction of node features from edge evidence, use of scalar edge weights inside message passing, normalization under species-level distribution shift, and post-hoc decision calibration. We compare multilayer perceptrons, GraphSAGE, and GIN baselines in PyTorch Geometric, using mean, sum, and max edge-to-node aggregation, Batch Normalization, Layer Normalization, and a species-conditioned Layer Normalization variant. Results are reported over three seeds with ROC-AUC, micro-F1, calibrated micro-F1, expected calibration error, training time, memory use, and parameter count. Sum aggregation is consistently the strongest edge-to-node construction. GraphSAGE with sum-based features forms the best accuracy-cost trade-off, with Batch Normalization reaching the highest ROC-AUC and conditional Layer Normalization retaining stronger fixed-threshold behavior. Per-label temperature scaling and per-label thresholds substantially improve multi-label decision quality with negligible change in ROC AUC, while light label-correlation smoothing yields small additional gains. The resulting protocol provides a compact, reproducible baseline for large edge-attributed biological graph settings. Together, these findings give practitioners clear default choices for feature construction, normalization, and decision calibration on edge-attributed protein graphs.

Energy efficiency of air-to-water heat pumps at different operating modes in residential heating systems - case study in Bosnia and Herzegovina

Krsto Batinić — 2026-09-14

This paper presents the possibilities of using air-water heat pumps for heating residential buildings. Description of the most
important parts of the air-water heat pumps, as well as the basic guidelines for design and implementation, are shown in this paper. Practical example with heat pumps settings and measurements of capacity and Coefficient of Performance (COP) and Seasonal Coefficient of Performance (SCOP) are described in this paper. All research regarding air-to-water heat pumps of different operating modes was conducted at a real facility in Pazarić, location near Sarajevo, Bosnia and Herzegovina. In addition, the energy efficiency class was determined on the considered example of the heat pump. Based on these measurements, calculation and by comparison, important conclusions were obtained regarding the use of observed heat pump, control and maintenance of such systems in terms of increasing energy efficiency related to indoor comfort in the residential building.