Manual essay correction methods can be time-consuming and hinder overall assessment efficiency. This study develops an Automated Essay Scoring (AES) system to address the inefficiencies of manual assessment, particularly for handwritten math exams. The proposed image-based AES system utilizes an optimized Two Dimensional Convolutional Neural Network (2D-CNN) approach to achieve faster and more accurate scoring. To facilitate this process, answer sheets have been pre-annotated with templates for easier identification during scanning. Initially, 40% of the answer sheets undergo manual evaluation to establish the ground truth for assessment, while the remaining 60% are used to train the CNN model. The performance of the previously developed system achieved an overall accuracy of 85% in classifying answer sheets based on their scores. This represents a substantial improvement compared to traditional methods of objective assessment. In conclusion, the AES system with CNN has the potential to significantly enhance the efficiency and accuracy of handwritten math essay assessments. This technology can save valuable time for educators and potentially enable more frequent or in-depth student feedback.
Keywords : Deep Learning; 2D-CNN; AES