Enhancing Handwritten Answer Assessment In Gujarati Language Education: A Comparative Study Of SVM, RNN-LSTM, And Bi-LSTM Approaches

Abstract

The automated assessment of handwritten student answers remains a significant challenge in the domain of educational technology, particularly for non-Latin scripts like Gujarati. This paper presents a comparative analysis of three machine learning algorithms: Support Vector Machine (SVM), Recurrent Neural Network with Long Short-Term Memory (RNN-LSTM), and Bidirectional Long Short-Term Memory (bi-LSTM), in their capacity to evaluate handwritten responses. We detail the process of digitising handwritten answers, the nuances of preprocessing for the Gujarati script, and the subsequent implementation of each algorithm. The bi-LSTM model demonstrated superior performance, with a 90% accuracy rate on training data and 80% on testing data, followed by RNN-LSTM and SVM. The study highlights the potential of using advanced machine learning models to automate the assessment process in regional language education systems and discusses the implications of such technologies in reducing educators' workloads and providing timely feedback. Future work will focus on refining script recognition, contextualising semantic analysis for Gujarati, and improving model adaptability to diverse handwriting styles.