Neural Networks Ensemble-Based IRT Parameter Estimation

Statistics methods are often used to estimate item parameter in item response theory (IRT). However, it has larger errors at small sample condition. The neural networks ensemble-based IRT parameter estimation method is put forward to solve this problem. The true values of item parameter are generated with computer simulation, examinees' response matrix is obtained based on two-parameter logistic model. The item difficulty p and discrimination r of classical test theory (CTT) are used as the inputs of generalized regression neural networks (GRNN). The simulated true values of IRT parameters are used as the outputs of GRNN. 30 neural networks are trained and the average of their outputs in the test phase is the estimate of IRT parameter. The results shows that neural networks ensemble could get better parameter estimation than statistics method and single neural network.