Predictive compilation results from student code smells with ELM

ABSTRACT

Code smells are not usually bugs, but it  refers to any symptom in the source code  that indicate weaknesses in design.In other word, It is directly related to the stability and robustness of the program. Therefore, new programmer should pay attention to these potential risks in particular，because a good programming style (avoiding code smell as much as possible) will lay a solid foundation for advanced programming. Over the past decade, we have observed significant growth of work in students' program data mining. However, the detailed course data is so rare that the majority of the studies focus on simplistic statistical analysis and are conducted within a small scale data set (less than 10000). In this work, we collected 17,854 code submissions in CS1, and use an open source platform SonarQube automatically inspect the code quality from the raw data. We then proposed an Extreme Learning Machine (ELM) based method to predict compilation results from processed data and educational platform data. In order to verify the accuracy of the forecast, we use the BP network as a comparison. The results show that our method is more accurate.