CNN-BASED CODE READABILITY CLASSIFICATION AND BUG LOCALIZATION IN PROGRAMMING FRAMEWORK

Abstract

Learning to program poses several challenges for students, particularly when it comes to writing readable code and effectively localizing bugs. Readability is a critical software quality attribute, aiding students in understanding their programs, identifying mistakes, and collaborating with others. However, the large volume and complexity of programming tasks often make it difficult for instructors to thoroughly assess the readability of every student's code. Similarly, bug localization remains a time-consuming and complex task, especially for students in the early stages of their coding education. Traditional methods for assessing code readability and localizing bugs tend to rely on simplistic metrics, which fail to fully capture the complexities of these phenomena. This research explores the potential of deep learning, specifically Convolutional Neural Networks (CNNs), to address these challenges. CNNs, originally designed for image processing, have shown promise in recognizing complex patterns within structured data, making them suitable for analyzing source code. With this, the researcher aims to create a framework on how the integration of a hybrid activation function model that combines ReLU (Rectified Linear Unit) and Leaky ReLU to enhance the performance of CNNs in these tasks. This will to contribute valuable insights into how deep learning models, particularly CNNs, can complement the learning experience of students, making coding education more efficient and effective.