https://scidar.kg.ac.rs/handle/123456789/8214 2026-07-22T11:24:46Z https://scidar.kg.ac.rs/handle/123456789/23196 Title: Истраживање компетенција за ВИ и употребе алата ВИ код студената учитељских студија Authors: Станковић, Марко; Тасић, Марија; Milenković, Aleksandar Editors: Svičević, Marina; Milenković, Aleksandar; Vučićević, Nemanja Abstract: Брза експанзија вештачке интелигенције (ВИ) довела је до развоја бројних алата применљивих у образовном контексту, при чему истраживања указују како на њихов значајан потенцијал, тако и на постојећа ограничења. Све веће присуство ВИ у образовању покренуло је важна питања у вези са компетенцијама студената и њиховом стварном употребом ВИ алата у академском контексту. Циљ овог истраживања био је да се испита однос између дигиталних и компетенција за ВИ код студената и њихове употребе алата ВИ у учењу. Истраживање је спроведено анкетним истраживањем на узорку од 82 студента учитељских студија у Србији. Подаци су прикупљени путем анонимног онлајн упитника. Факторском анализом издвојена су два фактора: дигиталне и компетенције за ВИ, као и употреба алата ВИ у учењу. Анализа поузданости показала је добру до високу унутрашњу конзистентност за оба фактора (Кронбах 𝛼 > 0.87), што указује на то да су коришћене скале поуздане и погодне за даљу анализу. Резултати показују да студенти процењују своје дигиталне и компетенције за ВИ као развијене, док је њихова стварна употреба алата ВИ у академском контексту знатно ређа. Ови налази указују на несклад између перципираних компетенција и стварне употребе, наглашавајући потребу за систематичнијом интеграцијом алата ВИ у образовању. Резултати имају импликације за образовну праксу, посебно у контексту подршке будућим учитељима у ефикасној примени технологија ВИ. 2026-01-01T00:00:00Z https://scidar.kg.ac.rs/handle/123456789/23190 Title: STUDENT QUESTIONS AS AN INDICATOR OF DIFFICULTIES IN SOLVING MATHEMATICAL COMPETITION PROBLEMS Authors: Grbović, Milica; Stojanović, Nenad; Milenković, Aleksandar; Vučićević, Nemanja Abstract: Mathematical competitions play an important role in fostering students’ interest in engaging more deeply with mathematics, as well as in developing a healthy sense of competition among peers. By solving more complex problems, students not only deepen their mathematical knowledge but also develop competencies for addressing and overcoming challenging situations. The aim of this study is to examine the types of challenges primary school students encounter when solving mathematical competition problems. The research was conducted through a thematic analysis of questions posed by students from grades 3 to 8 in the Šumadija District during municipal and regional competitions in the 2025/2026 school year. The results indicate that the identified challenges can be grouped into several categories: misunderstanding of the problem and its requirements; lack of familiarity with mathematical terminology and notation; difficulties in interpreting problem conditions; challenges related to the mental visualization of mathematical concepts and their relationships; absence of appropriate problem-solving strategies; and uncertainty regarding the correctness or meaning of the obtained solution. The paper presents representative examples for each category, along with an analysis of their distribution across different student age groups. The findings provide useful guidance for teachers in improving both regular and supplementary mathematics instruction, particularly in fostering students’ understanding of problems and their ability to select appropriate solution strategies, which may ultimately contribute to better performance in mathematical competitions. 2026-01-01T00:00:00Z https://scidar.kg.ac.rs/handle/123456789/23183 Title: Exploring Machine Learning Algorithms for Analysing Students' Attitudes Towards Distance Mathematics Learning Authors: Svičević, Marina; Milenković, Aleksandar; Krstić, Lazar; Pavković, Miloš Abstract: This study investigates the application of machine learning algorithms to analyse students' attitudes towards distance mathematics education, focusing on perceived effectiveness and students' ability to successfully learn and adopt mathematical content in an online setting. Data were collected from 1154 students at various educational levels using a 28- item Likert scale questionnaire on distance mathematics learning. An ML pipeline incorporating multiple data preparation techniques and machine learning algorithms was applied to two key prediction questions. Unlike predominantly descriptive or single model studies in this area, this approach evaluates both predictive performance and the stability of selected survey items across many model configurations, providing more robust and interpretable pedagogical insights. The results show that Recursive Feature Elimination was the most effective feature selection method, while Random Forest, Ridge Classifier and Categorical Naive Bayes achieved the strongest overall predictive performance across the two questions. These findings confirm the value of combining feature selection techniques and machine learning algorithms to derive robust and interpretable insights from educational survey data, while also highlighting the importance of well- structured distance learning strategies in mathematics education. The methodology is readily adaptable to other academic disciplines, providing educators with a data- driven framework for improving the design and effectiveness of online learning. 2026-01-01T00:00:00Z https://scidar.kg.ac.rs/handle/123456789/23182 Title: Qlora Fine-tuning of Mistral-7b for Serbian High School Mathematics Competition Tasks Authors: Pavković, Miloš; Svičević, Marina; Milutinović, Aleksandar; Vučićević, Nemanja; Milenković, Aleksandar Abstract: This paper examines the extent to which QLoRA fine-tuning can improve the performance of the large language model Mistral-7B-Instruct-v0.3 on Serbian high school mathematics competition tasks. Based on a dataset of tasks in Serbian, a fine-tuned model, Math-SRB-Mistral-7B, was developed and compared with the base model. The responses were evaluated using Claude 3.7 Sonnet as a judge, according to multiple criteria, including final answer accuracy, logical coherence, explanation quality, and an aggregate score. The results suggest that the applied fine-tuning did not lead to improved performance; instead, the fine-tuned model achieved slightly lower scores across all evaluated dimensions. This finding suggests that parameter-efficient adaptation of general-purpose LLMs on small and challenging mathematical datasets does not necessarily result in better generalization to new tasks. At the same time, the results highlight the importance of multi-criteria evaluation in the analysis of mathematical reasoning generated by LLMs 2026-01-01T00:00:00Z