INTELLIGENT CODE. SMELL. DETECTION SYSTEM USING DEEP LEARNING

Abstract

As the software industry is growing day by day, the challenges of software development are also growing. One of the major challenges in software development is the ability of evolution of software itself as per increase its demand, increase the feature, enhancement on itself. The evolution of software itself is not possible or extremely challenging as the code base of software itself is not scalable, maintainable, reliable, testable etc. Such type of code base is commonly considered as low-quality code and software build from such code base is low quality software. In this research, the goal is to understand the code smells and its importance and develop the intelligent model for code smell detection using one of the most popular machine learning algorithms, LSTM(Long Short-Term Memory), a RNN based approach. The data for training the model has been prepared by collecting the open-source codebase and trained, validate and test the model. The types of code smell considered for this research are Magic Number, Complex Method, Long Identifier, Long Statement, Long Parameter List, Deficient Encapsulation, Unutilized Abstraction, Insufficient Modularization, Broken Hierarchy and Feature Envy. The experimental results show that in the best case, the model produces an accuracy of 91.17%, True Positives 92.34% and False Positives 8.84%.