A Hybrid Approach to Student Assessment: Integrating C Code Metrics with Deep Learning Models

Abstract

Traditional methods of assessing programming skills primarily rely on output accuracy and manual grading, which can be time-consuming, subjective, and limited in scope. This paper introduces a hybrid, automated approach for evaluating students’ programming capabilities by analyzing the structural and behavioral aspects of their C code submissions. The proposed method involves extracting a rich set of static features from the C programs written by students during assessments. These features include lines of code, presence of comments, number and types of variables, condition checking count, loop count, retention of input, well-labeled output, return statement usage, variable initialization, shorthand notation, and detection of syntax errors. These attributes are indicative of a student's coding style, logic formulation, and understanding of programming constructs. To predict the academic outcomes, an Artificial Neural Network (ANN) is employed, which takes the extracted features as input and outputs the predicted grade and percentage of the student. The model is trained and validated on a dataset of real student code submissions, demonstrating high accuracy and consistency in performance prediction. This approach not only reduces the burden on educators but also ensures fairness, objectivity, and scalability in grading. Furthermore, it provides detailed feedback on specific aspects of student coding behavior, helping instructors identify strengths and areas for improvement. By combining code analytics with deep learning, this system establishes a foundation for intelligent education systems that adapt to student learning patterns and performance. It paves the way for data-driven academic decision-making and personalized feedback mechanisms in programming education. The results affirm that integrating code metrics with neural networks can revolutionize how programming assessments are conducted in academic settings.