A model for predicting educational vulnerability by neural networks

The aim of this study as part of a doctoral thesis was to develope a model for predicting educational vulnerability of undergraduate students in engineering disciplines in short term period (by semester). Method was data mining by using neural network algorithm. The statistical population, including all "Term- student" from the first semester in academic year 1390-91 till the second semester of 1393-94 in three Iranian engineering universities (with a total of 53,422 records). The needed data have been conveyed into model by direct exploitation from MISs in all three universities. The results indicate that by using the available data in educational systems of universities and engaging the neural networks algorithm it is possible to make a prediction by more than 95 percent accuracy and reliability over 60, for semester result about each student. “Before semester`s GPA”, “total GPA”, “being odd or even semester”, “type of semester`s units” and “engaging in extra activities”, were identified as the most effective predictive variables.