Estimation of Examinees' Ability Through Computer Adaptive Testing Based on Neural Network Approach

Examinee’s knowledge is measured through exams. A key purpose of using an exam is to determine the proficiency level of each examinee based on his/her responses to the administered test. A main problem of traditional test is that the asked questions are not match to actual ability of examinees and doesn’t measure examinee’s proficiency accurately. Computer Adaptive Testing (CAT) has been developed to address this issue. In CAT,each examinee has to answer the questions that are tailored to his/her ability level. Some of the features such as test security, immediate score reporting, improved efficiency and measurement precision have increased popularity of CAT. It uses models of proficiency estimation such as Item Response Theory (IRT). It is a statistical method with theoretical foundation that is being widely used in the field of modern educational testing technology and psychological testing. However, this model has some drawbacks. IRT model relates the response of an examinee to a specific item to his/her ability level and characteristics of the item. But relationship between items characteristics and person’s skill are very complex and nonlinear. In addition, it relies on strict assumption and need a large amount of data to precise measurement. These limitations are the motivation behind this research to use other adaptive approach to estimate the proficiency level in the CAT. In this thesis, we proposed a novel solution based on Artificial Neural Network (ANN) to address the above mentioned limitations. The ANN with adaptive features is a suitable scheme for solving complex non-linear problems. In addition, it has the ability to learn and generalize. These strong potentials make it an appropriate method to measure proficiency level of examinees in CAT systems. This work has been organized in two phases. In the first phase, we use 3-PL (three parameter logistic) dichotomous and polytomous model of IRT to estimate examinees’ ability in adaptive testing. Statistical approaches such as maximum likelihood estimation method and Bayesian approach are used for this purpose. In the second phase, estimation of examinees’ ability has been obtained with multi-layer feed forward neural network with back propagation algorithm. Experiments have been repeated under different scenarios and results indicate the advantages of the proposed scheme by obtaining better accuracy in performance.

Preview PDF FSKTM_2010_13_F.pdf Download (233kB)

Actions (login required)

View Item