Abstract
Automated Programming Assessment (APA) has evolved into an essential and effective method of assisting students in the computer programming learning, also providing students with proper support of formative assessment. Assessment feedback assists students in moving forward when they are struggling with their programming codes. Several studies have proposed various APA approaches. However, current research highlights a significant limitations: struggle to deliver feedback that is both immediate and contextually relevant, delayed and irrelevant feedback often results in reduced motivation and diminished learning effectiveness. This underscores the growing demand for APA tools capable of providing timely and meaningful feedback. Current APA studies emphasize that effective feedback should be timely, clear, relevant, and provide guidance aligned with the assignment requirements. To overcome these limitations, this study proposes an automated feedback generation approach called MCA-FGen to support a static analysis of APA using a Multi-Connect Architecture (MCA) Associative Memory Neural Network technique to handle syntax and semantic programming errors that arise during program compilation. Prior to this, a primary dataset containing 104 potential syntax and semantic programming errors was created for use in the learning phase MCA. The implementation and evaluation of the MCA-FGen and its corrective feedback were carried out through Technical and Controlled Experiments. The evaluation of the MCA-FGen in the technical experiment is demonstrated exceptional retrieval performance, achieving 100% accuracy across the primary dataset of manually captured syntax and semantic Java errors. High similarity scores (99.95%) and minimal Hamming distances (0–6 pixels out of 8100) confirm the fidelity of reconstructed patterns, while consistent retrieval times (0.4549 seconds in average) indicate stable convergence behavior. The controlled experiment involved 55 students and employed a feedback form as the data collection instrument. The forms demonstrated excellent internal consistency, with all constructs achieving Cronbach's alpha values above 0.90. The controlled experiment demonstrated a high level of student satisfaction with the proposed MCA-FGen. The overall evaluation across all forms dimensions produced mean scores ranging from approximately 6.5 to 6.9 on a 7-point Likert scale, indicating consistently positive perceptions. Among the evaluated dimensions, Timing achieved the highest overall mean, followed by Feedback Characteristics, Feedback Perception, and EUCS. These findings validate the MCA-FGen’s ability to reliably recognize and reconstruct error patterns, supporting accurate feedback generation. Overall, the results provide strong empirical support for the MCA-FGen’s core functionality while highlighting avenues for optimization in real-time educational contexts.