Evaluating Intelligent Methods for Software Risk Prediction: An Empirical Analysis

Abstract

Software development involves significant uncertainty due to unexpected events occurring at various stages of the software development lifecycle. As software size and complexity increase, the risk of project failures also rises. These unexpected events, known as software risks, stem from various factors throughout the development process. Effective risk management during the early phases is crucial to ensure a high-quality final product. Traditional risk assessments rely on human expertise and past experience, which can be subjective and less reliable. This study employs machine learning methods to predict software risks using historical data, aiming for early and accurate risk detection. Five machine learning models were evaluated alongside multiple feature selection techniques to improve prediction accuracy. Experiments were conducted using publicly available software risk datasets. Results show Support Vector Machine (SVM) model achieved highest classification accuracy of around 80%. Among feature selection methods, Mutual Information demonstrated superior performance across evaluated models, enhancing effectiveness of risk prediction.