Developing a Not-Reached Adaptive Test with K-Nearest Neighbor Method to Evaluate Examinees' Ability

摘要

Missing data is an inherent feature of most surveys or assessments that involve human subjects. In a sensor-driven Computerized Adaptive Test (CAT), not reached item is a kind of missing data issue which causes serious ability estimation problem. Previous studies tried to resolve this issue from the perspective of scoring rule. This study utilized a K-Nearest Neighbor (KNN) solution based on data mining method to impute the ability estimation for unreached items. The results indicated that the predominant of KNN method was not obvious when the value of k was less than 10. While the number of neighbor was larger than 20, the performance of KNN method was apparently better than previous proposed methods. Overall, the results indicated that the data mining mechanism might provide a better solution for not reached item problem.