Context Aware Hybrid Intelligence for Adaptive Software Reliability Assessment and Failure Prediction

Abstract

The increasing complexity of dynamic software environments, where the workload variations, resource conditions, execution behavior, and historical failure patterns are continually changing the system performance, makes software reliability assessment and failure prediction increasingly more difficult. The conventional methods typically feature static representations of features and may fail to adapt to varying operational contexts. A Hybrid Computational Intelligence Model (HCIM) for context-aware software reliability assessment and failure prediction is presented in this paper. The proposed framework combines context-aware feature modeling, adaptive temporal state representation, a neuro-fuzzy reliability inference, and evolutionary multi-objective optimization. Adaptive reliability representations are built by combining heterogeneous software execution features and information about the context in which the programs are executed. The hybrid intelligence mechanism is used to estimate software reliability, taking into account nonlinear relationships and uncertainty, while the context-guided prediction component takes care of the identification of potentially failure-prone states. Performance of reliability estimation and failure prediction is compared with representative machine learning, deep learning and neuro-fuzzy approaches, exhibiting better results. The proposed HCIM shows the reliability assessment with MAE, RMSE, and R^2 as 0.038, 0.055, and 0.968, respectively, and has an accuracy of 95.76% and an AUC of 0.972 for failure prediction. The results also show an increase in robustness in different working conditions. The proposed framework is shown to be effective and adaptive solution for intelligent software reliability management according to the results obtained.