In recent years, an array of studies has focused on ‘translationese’ (i.e., unique features that manifest in translated texts, causing second language (L2) writings to be similar to translated texts but different from native language (L1) writings). This intriguing linguistic pattern has motivated scholars to investigate potential markers for predicting the divergence of L1 and L2 texts. This study builds on this work, evaluating the feature importance ranking of specific translationese markers, including standardized type-token ratio (STTR), mean sentence length, bottom-frequency words, connectives, and n-grams. A random forest model was used to compare these markers in L1 and L2 academic journal article abstracts, providing a robust quantitative analysis. We further examined the consistency of these markers across different academic disciplines. Our results indicate that bottom-frequency words are the most reliable markers across disciplines, whereas connectives show the least consistency. Interestingly, we identified three-word lexical bundles as discipline-specific markers. These findings present several implications and open new avenues for future research into translationese in L2 writing.