Enhancing DevOps Pipelines With Viral Spread Optimization and Hybrid Deep Learning

Abstract

Modern software teams rely on Continuous Integration and Continuous Deployment (CI/CD) pipelines to automate testing and delivery. However, these pipelines often waste resources by running unnecessary tests and failing to predict which builds will fail. This study presents a new framework that addresses three key challenges: (1) deciding which tests to run first, (2) predicting which builds will fail, and (3) allocating computing resources efficiently. The framework combines Viral Spread Optimization (VSO) a new algorithm inspired by how viruses spread through populations with a hybrid deep learning model that pairs Bidirectional Long Short-Term Memory networks with Support Vector Machines (BiLSTM–SVM). VSO treats high-priority tests like "infections" that spread influence to similar tests, while the BiLSTM–SVM learns patterns from historical test data to predict failures. An ensemble module combining multiple classifiers further improves prediction reliability. Experiments on two large public datasets (TravisTorrent and CI-Datasets) showed that the framework detected 89.6% of faults, reduced test execution time by 21.5%, and cut cloud infrastructure costs by approximately 18%. These results represent significant improvements over existing methods, offering practical benefits for software teams managing large-scale automated pipelines.