A Lexical Model for Automated Essay Evaluation System Using Ensemble Learning Techniques

Automated Essay Scoring (AES) systems have become an attractive alternative to manual scoring due to their cost-saving benefits. This paper assesses the feasibility of such systems by conducting a study using a custom-built machine learning model, which employs the Random Forest Classifier to predict essay grades. Among the features used to train the model, individual and unique word counts emerged as the most effective in influencing the ratings. However, excessive reliance on these features has often been criticized and labeled as reductive. Regarding practical applications, the AES system developed in this study may be better suited for making estimations rather than serving as a substitute for human reviewers. In the future, improvements could focus on enabling the model to analyze higher-level features, such as flow and style.