Automated Grading System of handwritten Answer Sheets
The project proposed an automated system for grading handwritten answer sheets with the help of machine learning algorithm. For this purpose, 250 dataset samples were collected from students at Prince Mohammad Bin Fahd University. All the answer sheets were scanned separately through a portable scanner wherein, the scanner identified all the alphabetical and numerical answers written in each answer box and were stored as black and white images. After scanning each answer, the scanner performed segmentation to separate the questions from the answers written in each box. The segmentation procedure divided the images into more comprehensive divisions and procured more relevant data. Each character and digit answers were extracted using segmentation to generate parameters for testing and validation. The scanned images were fed through a machine learning classifier known as convolutional neural network (CNN) for the handwritten recognition purpose. Majority of the dataset samples collected was used to train the machine learning program and the rest was used for testing purposes. The system performed a high accuracy, resulting in a testing accuracy of 92.86%. The prototype indicated the feasibility of successfully building a system that could enable teachers to focus on necessary tasks and save time. Thus, the system can be considered as a fast and an open response offline grading system that could be used in universities, schools and colleges by the instructors to automatically grade the answer sheets of students faster and effectively.