Evaluating subjective papers manually is a challenging and labor-intensive endeavor. Key obstacles in using Artificial Intelligence (AI) for analyzing subjective papers include inadequate understanding and acceptance of data. Although numerous computer science approaches have attempted to score student answers, these generally rely on basic counts or specific words. Additionally, there is a notable scarcity of well-curated datasets for such tasks. This paper introduces a cutting-edge method that employs a variety of machine learning and natural language processing techniques, alongside tools like Wordnet, Word2vec, Word Mover’s Distance (WMD), Cosine Similarity, Multinomial Naive Bayes (MNB), and Term Frequency-Inverse Document Frequency (TF-IDF) to automate the evaluation of descriptive answers. Solution statements and keywords form the basis of the assessment, and a machine learning model is trained to predict the grades of these answers. Our findings indicate that WMD outperforms Cosine Similarity in overall effectiveness. With sufficient training, the machine learning model has the potential to operate independently. Our experiments have yielded an accuracy of 88% without the incorporation of the MNB model, and the error rate was further reduced by 1.3% through the integration of MNB.