Research on analysing Korean learners' vocabulary usage and selecting features to develop a proficiency prediction model

Objectives The objective is to analyse the vocabulary usage of Korean learners with the intention of developing a proficiency prediction model based on the findings. Methods To this end, we analysed vocabulary usage patterns using the learner morphological annotation spoken corpus provided by the National Language Institute of Korea, and trained and evaluated a prediction model using a random forest algorithm. Results The Random Forest-based prediction model demonstrated a 72.28% accuracy rate and a 16.33% mean absolute percentage error (MAPE), effectively reflecting learners' vocabulary usage and predicting mastery. In particular, the combination of variables including ‘total number of lexical implementations’, ‘total number of lexical types’, ‘number of sentences’, ‘number of content word implementations’, ‘number of content word types’, ‘number of n-class lexical types’, and ‘lexical density 1 and 2’ demonstrated the most favourable performance. Conclusions In conclusion, a random forest model comprising variables such as the number of lexical implementations, the number of types, and the number of sentences can accurately predict Korean language proficiency based on learners' lexical usage patterns. This can contribute to the development of automated learner assessment systems.