Milan Đorđević

Nije navedeno

Harnessing AI for Personalized Academic Major Recommendations An Application of Large Language Models in Education

Usman Durrani, Mustafa Akpinar, Asif Malik, Madeleine Togher, Milan Dordevic, Samer Aoudi

Harnessing AI for Personalized Academic Major Recommendations An Application of Large Language Models in Education | IEEE Conference Publication | IEEE Xplore

In the domain of educational counseling, the utilization of Large Language Models (LLMs) in conjunction with sophisticated AI technologies, such as embeddings and vector databases, introduces a groundbreaking methodology for advising students on their academic specializations. This study explores the complexities associated with conventional recommendation systems, which include issues like data imbalance and insufficient contextual awareness etc. By adopting a FewShot Learning framework, we harness the flexibility of LLMs to identify critical factors related to students’ interests and competencies. Our approach enables the dynamic extraction of important contextual information, thereby enhancing the predictive efficacy of the models. Through comprehensive experimentation with varied student datasets, we will experiment on if our AI system outperforms traditional recommendation techniques implemented by academic advisors and if it can yield detailed analyses customized to individual student profiles. This pioneering strategy holds considerable potential for advancing academic advising and assisting students in making well-informed choices regarding their educational trajectories.

Artificial Intelligence in Education, Large Language Models Personalized Academic Advising, Educational Recommender Systems, AI-Powered Recommendations

Nije navedeno

Nije navedeno

Assessing the Effectiveness of Large Language Models in Predicting Student Dropout Rates

Usman Durrani; Mustafa Akpinar; Madeleine Togher; Asif Malik; Milan Dordevic; Samer Aoudi

Assessing the Effectiveness of Large Language Models in Predicting Student Dropout Rates | IEEE Conference Publication | IEEE Xplore

Leveraging Large Language Models (LLMs) in educational analytics introduces promising and innovative approaches, particularly in predicting student dropout rates. One of these new approaches addresses the challenges of predicting student dropout rates. This study investigates the integration of advanced artificial intelligence (AI) tools, including OpenAI LLMs, HuggingFace embeddings, and FAISS vector databases, within a Few-Shot Learning framework. This study aims to identify subtle indicators of potential dropout risks by utilizing LLM’s contextual understanding and flexibility. Retrieval Augmented Generation (RAG) techniques were employed to improve the quality of contextual data provided to these models. Through extensive experiments with real-world data, we demonstrate that LLMs outperform traditional machine learning algorithms in predictive accuracy and provide valuable textual analyses of student data. This research highlights the potential of LLMs as powerful tools for dropout prediction in academic settings, contributing to ongoing efforts to reduce dropout rates.

Large Language Models, Student Dropout Forecasting, Few-Shot Learning, AI in Education, Contextual Data Analysis, Recommender Systems

Nije navedeno

Nije navedeno

Design, development, and evaluation of a mobile application for safety engineers

Suat Kasap, Ersin Elbasi, Milan Dordevic

Design, development, and evaluation of a mobile application for safety engineers

Hazards are the primary cause of occupational accidents, as well as occupational safety and health issues. Therefore, identifying potential hazards is critical to reducing the consequences of accidents. Risk assessment is a widely employed hazard analysis method that mitigates and monitors potential hazards in our everyday lives and occupational environments. Risk assessment and hazard analysis are observing, collecting data, and generating a written report. During this process, safety engineers manually and periodically control, identify, and assess potential hazards and risks. Utilizing a mobile application as a tool might significantly decrease the time and paperwork involved in this process. This paper explains the sequential processes involved in developing a mobile application designed for hazard analysis for safety engineers. This study comprehensively discusses creating and integrating mobile application features for hazard analysis, adhering to the Unified Modeling Language (UML) approach. The mobile application was developed by implementing a 10-step approach. Safety engineers from the region were interviewed to extract the knowledge and opinions of experts regarding the application’s effectiveness, requirements, and features. These interview results are used during the requirement gathering phase of the mobile application design and development. Data collection was facilitated by utilizing voice notes, photos, and videos, enabling users to engage in a more convenient alternative to manual note-taking with this mobile application. The mobile application will automatically generate a report once the safety engineer completes the risk assessment.

safety engineering; occupational safety and health; mobile application; unified modeling language (UML); occupational safety and health policy; risk assessment; hazard analysis

Nije navedeno

Nije navedeno

An Accurate Brain Tumor Segmentation using Deep Learning

Asad Safi, Milan Dordevic, Reem Atassi, Fuad AlHosban

An Accurate Brain Tumor Segmentation using Deep Learning | IEEE Conference Publication | IEEE Xplore

Gliomas, which can appear in different sizes, and locations, and with scattered boundaries, pose a significant challenge as an aggressive type of tumor. Convolutional Neural Networks (CNNs), one of the most effective deep learning approaches for image analysis problems, have been utilized to develop an automatic deep learning-based 2D-CNN model for brain tumor segmentation in this study. The architecture of the model was designed to be deeper by using small convolution filters (3x3). Additionally, the number of convolutional layers was increased to 16 for the HGG model and 13 for the LGG model to facilitate efficient learning of complex features from large datasets and achieve better results. Fine tuning among the dataset and hyperparameters was employed to obtain the results. The pre-processing for this model includes the generation of a brain pipeline, intensity normalization, bias correction, and data augmentation. The proposed method’s performance was evaluated using the BRATS-2015 dataset and the Dice Similarity Coefficient (DSC). Our method achieved DSC scores of 0.79, 0.77, and 0.78 for complete, core, and enhanced tumor regions, respectively. These results are comparable to other 2D CNN architecture-based methods.

Deep Learning, Neural Network, Image Analysis

Nije navedeno

Nije navedeno

AI-Enabled Assessment of Roadway Integrity: Forecasting Bitumen Deformation and Road Stability Throughout the Lifecycle Under Traffic Impact

Sabahudin Vrtagic, Milan Dordevic, Fatih Dogan, Muhammed Codur, Mario Hoxha, Edis Softic

AI-Enabled Assessment of Roadway Integrity: Forecasting Bitumen Deformation and Road Stability Throughout the Lifecycle Under Traffic Impact - TRID

Asphalt-paved Road junctions frequently encounter deformation and degradation challenges due to heavy vehicular traffic and varying climatic conditions, such as temperature fluctuations and precipitation. This study employs a multifaceted approach, incorporating a Multilayer Perceptron (MLP) model, ancillary machine learning techniques, and optimization methodologies, to address these challenges effectively. The primary objectives are the prediction and analysis of pavement deformation, the optimization of maintenance strategies, and the evaluation of road effectiveness. Our findings underscore the substantial contribution of heavy vehicles to road erosion and the profound impact of vehicular retention and braking at intersections. A Multilayer Perceptron (MLP) model is utilized to simulate future pavement degradation accurately at a specific intersection, leveraging real-time traffic flow data. This approach showcases the advantages of using real-world traffic data to model the lifecycle of asphalt dependencies dynamically at the intersection level. Mitigation of road deterioration is proposed via controlled traffic flow and optimization of relevant parameters, such as minimization of intersection wait times. The integration of machine learning substantially enhances road conditions and reduces vehicular waiting times at intersections. The implementation of this study's findings in pavement design and preservation practices could enable transportation authorities to improve road safety, reduce maintenance costs, and decrease the incidence of road accidents. Overall, this paper presents a comprehensive approach towards sustainable and efficient road infrastructure management, highlighting the potential of AI in tackling pressing infrastructure challenges.

Asphalt pavements; Backpropagation; Deformation; Intersections; Machine learning; Predictive models; Traffic loads

Nije navedeno

Nije navedeno

Development of Integrated Linear Programming Fuzzy-Rough MCDM Model for Production Optimization

Milan Dordevic, Rade Tešić, Srdjan Todorović, Miloš Jokić, Dillip Kumar Das, Željko Stević, Sabahudin Vrtagic

Development of Integrated Linear Programming Fuzzy-Rough MCDM Model for Production Optimization

One of the most common tools for achieving optimization and adequate production process management is linear programming (LP) in various forms. However, there are specific cases of the application of linear programming when production optimization implies several potential solutions instead of one. Exactly such a problem is solved in this paper, which integrates linear programming and a Multi-Criteria Decision-Making (MCDM) model. First, linear programming was applied to optimize production and several potential solutions lying on the line segment AB were obtained. A list of criteria was created and evaluated using the Improved Fuzzy Stepwise Weight Assessment Ratio Analysis (IMF SWARA). To obtain the final solution, a novel Rough compromise ranking of alternatives from distance to ideal solution (R-CRADIS) method was developed and verified through comparative analysis. The results show that the integration of linear programming and a Fuzzy-Rough MCDM model can be an exceptional solution for solving specific optimization problems.

linear programming; IMF SWARA; production optimization; Rough CRADIS

Nije navedeno

Nije navedeno

A New Data Fusion Model for Medical Image Encryption in IoT Environment

Reem Atassi, Fuad Alhosban, Milan Dordevic

A New Data Fusion Model for Medical Image Encryption in IoT Environment - United Arab Emirates - Ministry of Health and Prevention

An improvement of the Internet of Things (IoT) was forecast for changing the healthcare industry and is generating the increase of the Internet of Medical Things (IoMT). The IoT revolution was surpassed the present-day human service with promise social prospects, mechanical, and financial. During this condition, it can be essential for framing an effectual approach for guaranteeing the safety and reliability of t patient’s symptomatic information which are transmitted and received in IoT criteria. This study introduces a new data fusion model in IoT environment. The proposed model is called SSOECC-MIC model focuses on the design of effective encryption scheme with optimal key generation process for IoT environment. To achieve this, the SSOECC-MIC model designs an ECC model for the encryption and decryption of medical images effectively. To further improve the security performance of the ECC model, the optimal key generation process is carried out by the use of swallow swarm optimization (SSO) algorithm. For examining the enhanced performance of the SSOECC-MIC model, a wide ranging experimental analysis is carried out. The experimental outcomes reported the betterment of the SSOECC-MIC model over recent models.

Security, Data Fusion, Internet of Things, Healthcare, Medical images, Encryption, Key generation

Nije navedeno