Unveiling Hidden Patterns: Clustering Algorithms on C Code embedding
M Spoorthi, Richa Vivek Savant, Samarth Seshadri, Nakka Narmada, Peeta Basa Pati · 2024
Automatic grading can help streamline assessment by using advanced algorithms to evaluate embedded C programs swiftly and objectively. This technology ensures standardized evaluations, providing quick and objective feedback on programming proficiency, enabling educators and employers to gauge participants’ skills accurately. This paper proposes leveraging the application of machine learning clustering methods on embedding of C programs that have been examined by subject matter experts (SMEs) to uncover hidden patterns in the input vectors. A variety of clustering techniques are applied at the score level to capture the patterns in the data and analyze if they lie at par with the SMEs’ evaluations. Machine learning models can at best capture clusters at the questions level, i.e., it can capture similarities amongst text, but cannot comprehend the logic behind it and classify the types of errors in code solutions. K-means clustering proves to provide the best compact clusters at a question level of the chosen dataset with biased data distribution in the markings.