Modelling techniques applied to Automated Essay Grading: An Artificial Neural Network Backpropagation approach

Utilizing the Backpropagation method within Artificial Neural Network (ANN), Automated Essay Grading (AEG) has demonstrated notable efficacy in the evaluation of responses characterized by their open-ended and short-answer nature. The application of ANNB has proven to be a successful approach in accurately appraising and grading such types of written assessments. Despite advances in feature extraction and system evaluation, a research gap exists in adopting the ANNB approach to tackle the short AEG problem, leveraging its known capabilities in handling complex tasks. The dataset comprises 900 sets, with 582 subsets allocated for training and 318 for untrained data of technology short essays. Both datasets include actual marks assigned by subject matter experts to each short essay answer, ensuring a robust foundation for our analysis. The correlation coefficient has been employed to gauge the alignment between the predicted and actual marks of the models. Additionally, two error measures, namely Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE), have been utilized to assess the effectiveness of the developed model. This paper contributes novel insights to the discourse on AEG of short essays, aiming to bridge the identified research gap. Notably, there is a scarcity of research specifically exploring Backpropagation's application in the context of AEG, a void our paper seeks to fill by providing valuable insights into its relevance within this domain. While the reported metrics attest to the exceptional performance of the model, a critical discussion is warranted to examine potential outliers, sensitivity to variations, and real-world implications of observed errors.