Using Compiler Errors Messages to Feedback High School Students Through Machine Learning Methods

Teaching programming is essential for science and technology development in any country. Studies indicate high failure rates in programming subjects, which often lead to frustration or discouragement among students when they encounter coding and compiler error message. This is due to the language and interpretation of such error messages. Several efforts have been made to improve this situation, using neural networks or machine learning techniques to either fix the compiler error or give a more understandable error message. In this work, error-based learning is considered, and instead of fixing errors in the code, compiler error messages are used to provide feedback on syntax errors to programming students. The feedback consists of four components: a translation of the message to Spanish, syntax information about the language item, the error’s relation to possible causes and a reference to relevant topics for review. All the given information is intended to help students understand the error, allowing them to rewrite the code and compile it successfully. To achieve that goal, supervised learning was used to build a classifier using the compiler error messages. A set of documents was generated by injecting errors into model programs, thereafter, labelled according to the type of syntax error. The Machine Learning algorithms used were Decision Tree, Support Vector Machine, Random Forest, Multi-layer Perceptron and K-Nearest Neighbors. These classifiers were trained using 80% of the documents and evaluated with the remaining 20%, achieving an accuracy of over 90% for making new predictions and providing feedback.