A hybrid model based on learning automata and cuckoo search for optimizing test item selection in computerized adaptive testing

Computerized adaptive testing (CAT) has become one of the most effective methods of providing a customized approach to assessment. Nevertheless, the identification of student ability and the choice of appropriate test items is still a problem. In this paper, a new design of CAT strategy is proposed based on reinforcement learning, machine learning, and multi-objective optimization. A learning automaton is used to feed a neural network with the ability to predict a student's ability based on their response history. A multi-objective cuckoo search algorithm then decides the test strategy, between content coverage and level compliance. Compared with the traditional CAT methods, our approach gives better ability estimates and selects test items that are most appropriate for each student. The findings of the study show that the efficiency, accuracy and fairness of the tests have improved through experimentation.