University of Twente Student Theses

Login

What is the comparative accuracy of IRT, linear regression, and XGBoost in predicting students' performance on different versions of an exam?

Urbanski, Merlin Torben (2023) What is the comparative accuracy of IRT, linear regression, and XGBoost in predicting students' performance on different versions of an exam?

[img] PDF
569kB
Abstract:Using multiple exam versions to avoid cheating in the form of copying and teamwork is common practice at universities and schools. To ensure fairness during the grading process in many cases the versions share common items that are then used for the harmonization of the two versions. In this thesis, a comparative analysis was conducted to assess the accuracy of three harmonization methods: Item Response Theory (IRT), Linear Regression, and XGBoost. A fictional scenario was created involving 200 students who took an exam with two versions containing 20 exclusive items and 10 shared items. Two different data sets, one simulated and the other based on real data, were used for the analysis. The results indicated that IRT performed the best in terms of accuracy for both data sets. Linear regression was found to be the second most accurate method, while XGBoost showed comparatively lower accuracy. The methods used in this study were much simpler compared to other research, so it is important to interpret the results with caution.
Item Type:Essay (Bachelor)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:77 psychology
Programme:Psychology BSc (56604)
Link to this item:https://purl.utwente.nl/essays/95693
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page