The use of decision tree algorithm in computerized adaptive tests based on cognitive diagnostic models

Cognitive Diagnostic Models (CDM) and Computerized Adaptive Testing (CAT) remain topics of ongoing development and popularity in the field of educational measurement. This study proposes a method for adaptive tests based on Cognitive Diagnosis Models (CDM-CAT) that incorporates starting, item slection, and termination rules grounded in decision tree algorithms. Simulations were conducted to compare the proposed method with traditional CDM-CAT applications under varying conditions of item pool size, item discrimination levels, and test length, assessing accuracy of latent class and skill estimation. Additionally, similar comparisons were made using real student responses. Data derivation and analysis were carried out using various packages in the R programming language. Results show that the decision tree-based method achieved higher rates of successful class and skill estimation compared to traditional methods when using real data. Moreover, under simulation conditions with test lengths of 18 and 24 items, the decision tree-based method demonstrated better performance. In contrast, traditional applications employing the Global Discrimination Index (GDI) progression rule yielded higher successful classification rates in conditions utilizing 12-item test lengths.