Writing in L2 is both crucial and difficult for teachers and students since most of the assessment in higher education is in written form. The production of texts as well as providing feedback to them requires time and effort on both sides. For this reason, prediction of future L2 writing performance in advance may prove quite useful both for teachers and students and the information obtained can be used for failure prevention. In this respect, the present study aimed to train a machine learning model which used demographic and psychometric data to predict end-of-term L2 writing performance on a Pass/Fail basis. Secondly, the study aimed to find out if synthetically augmenting available data would increase prediction accuracy. The data were collected in the beginning of the semester from 102 students of English Language Teaching in Turkey. L2 writing performance was measured at the end of the semester. The findings indicated that Instance-Bases Learning with Parameter K algorithm could accurately predict L2 writing performance in advance. Moreover, artificial augmentation of the data was seen to increase prediction accuracy. The findings bear implications for the effective prevention of student failure by detecting poor performance in advance.