Recognizing Algorithms Using Roles of Variables, Language Constructs and Software Metrics: A Machine Learning Approach

Roles of Variables (RoV) are concepts that describe the behavior and usage of variables in computer programs. RoV have originally been introduced to help novices learn programming. Algorithm Recognition (AR) is a subfield of program comprehension, where the problem is to identify algorithms from the source code. AR covers recognizing different algorithms that carry out different computational task, as well as different types of algorithms that perform the same task. The main application of AR is in automatic assessment of students' programs to verify that the program implements the required algorithm. This thesis investigates the applicability and usefulness of RoV in AR. The idea is to analyze different implementations of basic algorithms to see whether RoV appear in those algorithms is such a way that they can be distinguished on this basis. In addition to RoV, other distinctive characteristics that can be used in the recognition process are also identified from the algorithms. These characteristics are related to language constructs and various software metrics. Based on the results of these analyses, a method for AR is introduced and a tool for automatic algorithm recognition is developed. Two experiments on sorting algorithms are carried out to illustrate the feasibility of the AR method. In the first experiment, sorting algorithms are recognized using a manually constructed decision tree. The second experiment uses the C4.5 algorithm to construct the decision tree. The results of the experiments (86% and 97.1% correctly recognized algorithms respectively) show the applicability of RoV and the other characteristics in AR problem. Moreover, the performance of the automatically constructed decision tree demonstrates that machine learning techniques are also suitable for AR task. and was partially funded by the Academy of Finland. I would like to thank my supervisor, Professor Lauri Malmi, for providing me the opportunity to do this work. I highly appreciate his valuable input, feedback and suggestions during the project. I am sincerely thankful to him. I am also grateful to my instructor, Docent Ari Korhonen, for his insightful comments, helpful ideas and great discussions. Finally, I would like to express my greatest gratitude to my family; my wife, Elmira, my daughter Ava and my son Arad, for their love, understanding and encouragement, as well as my parents for their support over the years. This paper discusses roles of variables, their application to algorithm recognition and introduces a method for algorithm recognition. It also presents an extensive literature …