DNN-based Software Reliability Model for Fault Prediction and Optimal Release Time Determination

The accurate prediction of both detected and corrected faults is crucial for enhancing software reliability and determining optimal release times. Traditional Software Reliability Growth Models (SRGMs) often focus on either fault detection or correction, potentially overlooking the comprehensive view needed for effective software maintenance. This paper introduces a Dense Neural Network (DNN)-based model that predicts both detected and corrected faults using data from the initial testing phase. The proposed model adopted a simpler architecture to reduce computational overhead and minimize time complexity, making it suitable for real-world applications. By incorporating logarithmic encoding, the model effectively manages missing data and performs well with smaller datasets, which are common in early testing stages. The proposed model is compared with existing approaches, demonstrating superior results across multiple datasets. This comparative analysis highlights the model's enhanced predictive accuracy, computational efficiency, and less time complexity. Additionally, the predicted faults are used to determine the optimal release time, based on the customer's reliability requirements and the minimum cost necessary to achieve that reliability. By offering a more comprehensive and accurate prediction of software reliability, this model provides a practical solution for software development teams, facilitating better decision-making in testing, maintenance, and release planning.