Predicting item difficulty in a language test with an adaptive neuro fuzzy inference system

This study reports a novel application of the Adaptive Neuro Fuzzy Inference Systems (ANFIS) to a second language listening test, and compares it with path modeling of observed variables. Seven variables were defined and hypothesized to influence the primary dependent variable, test item difficulty. Next, a matrix of these eight variables was developed and subjected to ANFIS and path modeling. ANFIS analysis found stronger effects for several of the seven explanatory variables. Path modeling captured some of the same effects through a mediating variable, test section, which captures aggregate differences across different subsections of the test. In general, neurofuzzy models (NFMs) appear to be a promising tool in language and educational assessment.