Scientific Journal of Astana IT University
Not a member yet
250 research outputs found
Sort by
DEVELOPMENT OF IMAGE CAPTION GENERATION HYBRID MODEL
Get PDFThis study presents a hybrid model for image captioning using a VGG16 convolutional neural network (CNN) for feature extraction and a long short-term memory (LSTM) network for sequential text generation. The proposed architecture addresses the challenges of producing semantically rich and syntactically accurate signatures, especially in languages with limited training data. The model effectively bridges the semantic gap between visual and textual modalities by utilizing pre-trained weights and a robust encoding-decoding system. Experimental results on a dataset of road signs in Kazakhstan show a significant improvement in inscription quality as measured by BLEU and METEOR metrics. The model achieved a maximum METEOR score of 0.9985, indicating high semantic accuracy, and BLEU-1 and BLEU-2 scores of 0.67 and 0.64, respectively, highlighting the model's ability to generate relevant and coherent captions. These findings underscore the model's potential applications in multimodal systems and assistive technologies. Using a pre-trained CNN model (VGG16), we can efficiently encode visual information by extracting high-level features from images. This approach is particularly useful for tasks that require consideration of the semantics of images, such as road sign recognition. The second LSTM model, as a sequence-oriented architecture, is well-suited for text generation, as it effectively considers the context and previous words in a sequence. These models can be integrated into systems requiring the analysis and description of visual information, such as autonomous vehicles or driver assistance systems. In conclusion, the proposed model demonstrates high potential for image caption generation tasks, especially in resource-constrained environments and for specialized datasets
CONSTRACTION OF DISTRIBUTION MODELS OF THE UNIVERSITY EDUCATIONAL WORK VOLUME
Get PDFWith the advent of new time requirements for the quality of educational services, which is influenced by management in functioning business processes, existing research in the field of resource allocation in the management of complex processes, namely the calculation of the teaching load of university teaching staff, was studied. The purpose of the research in this article is to develop functional and mathematical distribution models of the university educational work volume, as well as an algorithm for optimizing the generation of educational flows and initialization of academic groups, taking into account the specifics of disciplines and classroom fund. The algorithm is based on the construction of all business processes implemented during the formation of educational streams and groups. The functional model described for the process of distributing the volume of educational work includes the definition of the main functions, their relationships, input and output data, as well as the criteria and restrictions that govern this process. The mathematical model is based on the representation of all types of educational work of departments of educational programs as a discrete set of resources that must be distributed between educational departments in accordance with the assumptions and restrictions accepted at the university. Data mining and operations research techniques were used to write the functional model. Empirical and quantitative methods were used to write a mathematical model. Thus, a new methodology has been developed for solving complex optimization problems that arise when modeling and optimizing the distribution of the volume of educational work of a university. It should be noted that comparative experiments under labor-intensive and time-limited conditions confirm the effectiveness of this technique in solving problems of distributing the amount of educational work among departments of educational programs, which in turn contributes to the implementation of high-quality software
COMPARATIVE ANALYSIS OF VARIOUS FORECAST MODELS OF ELECTRICITY CONSUMPTION IN SMART BUILDINGS
Get PDFThe rapidly growing field of smart building technology depends heavily on accurate electricity consumption forecasting. By anticipating energy demands, building managers can optimize resource allocation, minimize waste, and enhance overall efficiency. This study provides a comprehensive comparative analysis of various models used to forecast electricity consumption in smart buildings, highlighting their strengths, limitations, and suitability for different use cases. The investigation focuses on three major categories of forecasting models: statistical methods, machine learning techniques, and hybrid approaches. Statistical models, such as the Moving Average Method, leverage historical data patterns to predict future trends. These models enable analysts to utilize predictive analytics, simulating real-world environments and helping them make more informed decisions. The study offers a detailed comparison of several predictive models applied to Internet of Things (IoT) data, with a particular emphasis on energy consumption in smart buildings. Among the short-term forecasting models examined are gradient-enhanced regressors (XGBoost), random forest (RF), and long short-term memory networks (LSTM). The performance of these models was evaluated based on prediction errors to identify the most accurate one. Time series, machine learning, and hybrid models used to predict energy consumption are considered and analyzed. The focus is on the accuracy of forecasts and their applicability in real-world conditions, considering factors such as climate change and data obtained from Internet of Things (IoT) sensors. The analysis shows that hybrid models combining machine learning and time series provide the best prediction accuracy over different time horizons. It also highlights the importance of integrating user behavior data and using IoT technologies to improve model accuracy. The results can be applied to create energy-efficient control systems in smart buildings and optimize energy consumption
NEW APPROACH TO ADDRESSING CLASS IMBALANCE IN MEDICAL DATASETS CONSIDERING SPECIFICS
Get PDFCurrently, the popularization of the integration of machine learning into the field of medicine for data processing and analysis is being traced, but at the same time difficulties such as class imbalance and noisy datasets arise. Due to the prevalence of the problem, there are already existing solutions, but in all of them there is an abstraction from the field of medicine, namely, gender, racial and other differences are not taken into account. It is this side of the problem that is solved in our resampling algorithm. A feature of our algorithm is the use of splitting the dataset by an important feature through the p-value of Spearman correlation, which helps to consider subgroups of observations without losing their unique characteristics and removing noise data using LOF and Z-score separately for minority and majority classes, respectively. Synthetic data is generated in a flexible way, adapting to the data set using algorithm parameters. Work is provided with both quantitative and nominative features. The algorithm was tested on datasets for heart attack, chronic kidney disease, and liver disease, and the Random Forest ensemble method was used to train the model. After applying this class balancing method, improvements were recorded on average in Accuracy by 36%, in AUC by 15-25%, in Precision by 39-42%, and in Recall by 21-37% compared with SMOTE, ADASYN algorithms and the data set before balancing. Applying the algorithm on medical data can improve the accuracy of the algorithm and reduce the loss of reliability compared to other resampling methods
METHODS OF INFORMATION SECURITY IN THE INTERNET OF THINGS (IOT) NETWORKS USING QUANTUM MACHINE LEARNING
Get PDFThe development of the Internet of Things (IoT) poses serious security challenges due to the vulnerability of devices and network connections. IoT devices often have limited computing resources, which makes it difficult to implement traditional security methods such as encryption and intrusion detection systems. In addition, the dynamic nature and high complexity of IoT networks create additional security challenges, requiring the development of new, more effective security methods. Traditional machine learning algorithms used to protect IoT networks have their limitations in terms of scalability and ability to effectively cope with large volumes of data, as well as new types of threats. These algorithms are often unable to quickly respond to anomalies, which significantly increases the risk of cyberattacks. In this regard, there is a need to find new solutions to improve the security of IoT networks.
This paper proposes a new approach to IoT security using quantum machine learning (QML), which combines the capabilities of quantum computing with machine learning algorithms to create more powerful models for detecting threats and anomalies in IoT networks. We analyze various QML algorithms, such as quantum support vector machines (QSVMs), quantum neural networks (QNNs), and quantum reinforcement learning (QRL), applied to solve security problems. Experiments conducted using the dataset confirm the effectiveness of quantum algorithms compared to traditional machine learning methods. The results show that QML models provide higher accuracy in detecting threats and anomalies, and significantly reduce the time spent on processing and training compared to classical methods. In conclusion, we argue that using QML to protect IoT networks can significantly improve their security and efficiency, opening up new prospects for further research in this area
DEVELOPMENT OF METHODOLOGY FOR GIS ASSIGNMENTS IN FLOOD MAPPING USING BLOOM’S TAXONOMY AND SCAFFOLDING APPROACH
Get PDFThis paper presents a study on designing and implementing a series of GIS assignments for an educational course on flood mapping, structured using Bloom’s Taxonomy and the scaffolding teaching method. Geographic Information Systems (GIS) education often involves the acquisition of complex technical skills, requiring a structured learning approach to ensure a progressive mastery of concepts. In this study, a sequence of practical assignments was developed at increasing levels of complexity corresponding to Bloom’s cognitive levels, from basic knowledge acquisition to higher-order evaluation tasks. The scaffolding approach was utilized to facilitate student learning, wherein extensive guidance was provided in early tasks and gradually removed in later ones as students gained competence. The research was conducted in an upper-level undergraduate course, “Methodology for Mapping Flood Emergency Areas”, at the Sarsen Amanzholov East Kazakhstan University, with 21 enrolled students. The assignments integrated real-world flood mapping scenarios using GIS tools such as ArcGIS Pro and QGIS, enabling students to apply theoretical knowledge in practical settings. Results from the study indicated that a structured, scaffolded approach significantly improved student performance and confidence in GIS skills. Quantitative analysis of assignment grades showed steady improvement as students progressed to more complex tasks, while qualitative feedback revealed high engagement and perceived learning value. The findings underscore the effectiveness of combining Bloom’s Taxonomy with scaffolded instruction in GIS education, providing a practical framework for curriculum design. This approach has the potential to enhance learning outcomes in technical subjects, particularly in geospatial analysis, and offers recommendations for educators on implementing scaffolded assignments effectively. Further research could explore long-term skill retention and the application of this methodology in other technical disciplines
PREDICTING DIABETES PROGRESSION USING AN ENSEMBLE OF CNN, RNN, AND LSTM MODELS
No full textThis article presents an integrated approach to predicting diabetes progression based on an ensemble of multiple deep neural network architectures. To enhance diagnostic accuracy and reliability, convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory (LSTM) models are jointly utilized within a clinical decision support framework. The optimal combination of their predictions is achieved through the Dirichlet ensemble method, which adaptively distributes weights among individual models according to their validation performance. Hyperparameter optimization using the Grid Search algorithm allows systematic selection of training parameters, network depth, activation functions, and regularization techniques, ensuring better convergence and reducing overfitting risks. The study involves a comprehensive data preprocessing pipeline, including normalization, balancing, and One-Hot Encoding of categorical features, to manage heterogeneous medical datasets and minimize the effect of missing or noisy information. Experimental evaluation demonstrates that the proposed ensemble model significantly outperforms individual CNN, RNN, and LSTM architectures in terms of accuracy, sensitivity, and stability, achieving improved generalization ability and robustness to data variability. This research emphasizes the potential of ensemble deep learning approaches to strengthen modern clinical decision support systems (CDSS). The developed framework enables more precise and interpretable diagnostic predictions, contributing to early diabetes detection and prevention strategies. Furthermore, the proposed methodology can be extended to other medical classification problems, providing a flexible and adaptive analytical tool for healthcare applications. The findings confirm that adaptive ensemble methods based on the Dirichlet distribution can serve as a foundation for reliable, transparent, and intelligent clinical decision-making in future healthcare systems
DEVELOPMENT OF A NEURAL NETWORK-BASED MODULE FOR FORECASTING ATMOSPHERIC POLLUTANT EMISSIONS
No full textThe prediction of emissions to air is a crucial and complex task for environmental monitoring and air quality management. Accurate forecasting is essential for the timely adoption of mitigation measures and for ensuring regulatory compliance. However, traditional statistical methods often perform inadequately because they poorly capture non-linear dependencies, intricate interactions between variables, and long-term temporal patterns, all of which ultimately decrease forecasting accuracy. The work presents an emission prediction software module based on a neural network with LSTM architecture. The input factors used were the concentrations of the main pollutants (NO, NO2, SO2, CO, solid particles) as well as meteorological indicators including air temperature, humidity and flow rate. Data provided by the operating enterprises, including 39,803 lines with increments of 20 minutes, were pre-processed: cleared from skips, normalized parameters and forming training sequences of 72 steps, corresponding to the daily interval. Additional exploration analysis was performed, which revealed the presence of expressed daily and weekly cycles, as well as correlations between weather conditions and concentrations of pollutants. The built model showed the ability to reproduce emission dynamics with acceptable accuracy, which is confirmed by MSE 0.87 and R2 0.86 values. The developed module is integrated into the current monitoring system and provides a user-friendly interface for building real-time forecasts. The results are consistent with current research, but the work is applied as a tool used in practical activities. In the future, it is planned to expand the set of factors and explore the possibilities of using ensemble architecture to improve the accuracy and robustness of forecasts
IMPACT OF LOSS FUNCTION ON SYNTHETIC BREAST ULTRASOUND IMAGE GENERATION
No full textThe BUSI (Breast Ultrasound Images) dataset is small and imbalanced, which limits the effective training of deep learning diagnostic models. Generative Adversarial Networks (GANs) offer a promising and increasingly popular solution for synthesizing realistic medical images to augment scarce training data and improve overall model generalization. This study investigates the impact of loss function selection in our previously published Deep Generative Adversarial Network with Wasserstein Gradient Penalty and Transfer Learning (DGAN-WP-TL). Two configurations were evaluated: one trained using Wasserstein GAN with Gradient Penalty (WGAN-GP) and another trained using Binary Cross-Entropy (BCE) loss. The experiments were conducted on the BUSI dataset with perceptual loss weights λ = 0.5, 3.0, 5.0, 7.0, and 10.0. Model performance was comprehensively assessed using Fréchet Inception Distance (FID), Kernel Inception Distance (KID), Learned Perceptual Image Patch Similarity (LPIPS), and Multi-Scale Structural Similarity Index (MS-SSIM). Results demonstrate that WGAN-GP consistently outperformed BCE across all λ values, generating images with higher fidelity, improved realism, and greater visual diversity. The superiority was most pronounced for λ = 3.0 and λ = 5.0, where WGAN-GP achieved the lowest KID and FID and the most balanced diversity–fidelity trade-off. The best-performing DGAN-WP-TL configuration (WGAN-GP, λ = 5.0) achieved KID = 0.14, FID = 179.42, LPIPS (fake–fake) = 0.49, and MS-SSIM (fake–fake) = 0.18. These results highlight the crucial role of loss function design in medical image synthesis. Overall, the study confirms that WGAN-GP provides superior image realism and variability, making it the preferred choice for high-quality, clinically relevant synthetic data generation, while BCE remains a lightweight and practical alternative for constrained computational environments
MODELING AND WEB-BASED VISUALIZATION OF FLOOD ZONES: A CASE STUDY OF THE BUKTYRMA RIVER SECTION
Get PDFThis study focuses on flood risk assessment in a vulnerable reach of the Buktyrma River basin, located in the East Kazakhstan region. The main goal was to develop and validate a modelling workflow that uses open-access data and software to simulate flood dynamics and visualize the results through an interactive GIS-based interface. The approach involved statistical analysis of 24 years of annual maximum discharge and water level data from gauging station to define a representative flood event. Terrain data were derived from the 30 m Copernicus Digital Elevation Model, which was used to construct the hydraulic geometry of the study area. Two-dimensional flood modeling was carried out in HEC-RAS 6.6, incorporating spatially differentiated Manning’s roughness values based on cadastral land-use classification maps.
The modeling results were verified using satellite imagery from Landsat 7 by calculating the Normalized Difference Water Index for the 2018 flood event, which had a 4 % exceedance probability. The comparison showed a high degree of agreement, indicating that the simulated flood zone overlapped 87 % with the NDWI-derived water mask, and the total inundation area differed by less than 2 %. Model outputs such as flood depth, flow velocity, and cross-sectional profiles were visualized, and the resulting flood map was uploaded within a web-GIS platform. The study demonstrates a transparent and cost-effective methodology that can be applied to other river basins in Kazakhstan, offering a practical tool for spatial planning, risk mitigation, and early warning systems based entirely on publicly available data and software