A Machine Learning-Based Reliability Assessment Model for Critical Software Systems

Service-oriented architecture (SOA) techniques are being increasingly used for developing critical applications, especially network-centric systems. While the SOA paradigm provides flexibility and agility to better respond to changing business requirements, the task of assessing the reliability of SOA-based systems is challenging, especially for composite services. However, deriving high confidence reliability estimates for mission- critical systems can require huge costs and time. This paper presents a reliability assessment and prediction model for SOA-based systems. The services are assumed to be realized with reuse and logical composition of components. The model uses AI reasoning techniques on dynamically collected failure data of each service and its components as one of the evidences together with results from random testing. Memory-Based Reasoning technique and Bayesian Belief Networks are used as reasoning tools to guide the prediction analysis. The least tested and "high usage " input subdomains are identified and necessary remedial actions are taken depending on the predicted results from the proposed model. The model is illustrated using a simulated case study based on a real-time dataset from the NASA software repository.