Composing Multiple Online Exams: The Bees Algorithm Solution

Abstract

Online education has gained increasing importance in recent years due to its flexibility and ability to cater to a diverse range of learners. The COVID-19 pandemic has further emphasized the significance of online education as a means to ensure continuous learning during crisis situations. With the disruption of traditional in-person exams, online examinations have become the new norm for universities worldwide. Among the popular formats for online tests are multiple-choice questions, which are drawn from a large question bank. However, creating online tests often involves meeting specific requirements, such as minimizing the overlap between exams, grouping related questions, and determining the desired difficulty level. The manual selection of questions from a sizable question bank while adhering to numerous constraints can be a laborious task. Additionally, traditional search methods that evaluate all possible solutions are impractical and time-consuming for such a complex problem. Consequently, approximate methods like metaheuristics are commonly employed to achieve satisfactory solutions within a reasonable timeframe. This research proposes the application of the Bees Algorithm (BA), a popular metaheuristic algorithm, to address the problem of generating online exams. The proposed solution entails creating multiple exam forms that align with the desired difficulty level specified by the educator, while considering other identified constraints. Through extensive testing and comparison with four rival methods, the BA demonstrates superior performance in achieving the primary objective of matching the desired difficulty level in most test cases, as required by the educator. Furthermore, the algorithm exhibits robustness, indicated by minimal standard deviation across all experiments, which suggests its ability to generalize, adapt, and be practically applicable in real-world scenarios. However, the algorithm does have limitations related to the number of successful solutions and the achieved overlap percentage. These limitations have also been thoroughly discussed and highlighted in this research.