Software Defect Prediction with Fuzzy Logic

Abstract

Finding software defects in software project modules is a complex process and highly uncertain in nature. Even though multiple intensive machine learning and deep learning models are available to predict defects, it is important to define and construct a simple model that applies the domain expert's knowledge and handle uncertainty in measurement of features. We developed a Mamdani Fuzzy Logic-based Software Defect Prediction model that accepts both traditional membership functions (Triangular, Trapezoidal, etc) and domain expert's custom membership function to predict software defects. To improve upon the Mamdani system, we implemented a simple Takagi Sugeno model that provided better predictions. We evaluated our fuzzy logic models using popular regression models like Multiple Linear Regression and Random Forest Regression.