Predictive Modeling of Software Behavior Using Machine Learning

Abstract

Predicting software behavior is crucial for verifying software properties and ensuring system reliability. Comprehending the behavior of complex software systems is particularly challenging because of characteristics such as concurrency and resource constraints, which lead to the interleaving of execution paths. Consequently, accurately predicting the behavior of such software may pose significant challenges. To address this challenge, we propose a Software Behavior Prediction (SBP) model that comprehensively captures and accurately predicts software behavior. By applying the machine learning concepts of Multi-label Classification (MLC) and Long Short-Term Memory (LSTM) networks, the SBP model enhances the accuracy and efficiency of identifying target states or software behaviors. Furthermore, we introduce a threshold determination method designed to improve the accuracy of MLC tasks. In practical applications, our SBP model can iteratively and thoroughly predict the states in the future moments of the execution paths, eliminating the need for analysis from the initial state of the software. This capability enables our approach to predict software behavior with higher accuracy and efficiency than the baseline models. The experimental results demonstrate the effectiveness of the SBP model in accurately predicting multiple concurrent behaviors within complex software systems. Additionally, the SBP model demonstrates improved resource efficiency compared to the baseline methods. These findings highlight the practical relevance and potential impact of the SBP model in real-world scenarios, offering promising avenues for enhancing software design techniques and improving system reliability.