Reinforcement learning framework for computerized adaptive testing using multi armed bandit approach

This research introduces a new framework based on the combination of deep learning and reinforcement learning techniques for computer-based adaptive testing that overcomes the limitations of traditional methods for test preparation. By effectively modeling the test taker’s response patterns and utilizing mathematical modeling techniques, our approach attempts to demonstrate optimal performance in estimating the test taker’s abilities and designing the test according to his abilities. Our proposed method is carried out in three main stages. In the first stage, a one-dimensional deep neural network is used to estimate the test taker’s ability level, which is performed periodically. In the second stage, a model based on the multi-armed bandit (MAB) problem is used to select the test strategy. In the mathematical model, the objectives are to consider the goals of content coverage and matching the test taker’s level. At the third stage, the updating of parameters and repetition of the test strategy determination operation is done through a reinforcement learning model. Through these integrated steps, our proposed method provides an effective approach to adaptive testing that is closely aligned with individual capabilities. The experimental results showed that our proposed method outperformed the comparative methods and achieved Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) values ​​of 1.53 and 1.21, which are significant compared to other methods. These results confirm the effectiveness and superiority of our approach in estimating abilities and designing tests that are tailored to the individual capabilities of respondents.