A Hybrid Machine Learning and Artificial Intelligence Framework for Intelligent Computing, Computer Vision, and Predictive Decision-Making in Next-Generation CSE Applications

Abstract

Contemporary computer science and engineering (CSE) systems increasingly combine multiple, architecturally distinct artificial intelligence paradigms within a single deployed pipeline: convolutional and transformer-based perception modules for computer vision, ensemble and gradient-boosted models for structured predictive decision-making, and, increasingly, neuro-symbolic components that reintroduce explicit rule-based reasoning alongside learned representations. This paper reviews the technical foundations underlying this hybrid design pattern, tracing the evolution of deep learning architectures for computer vision from convolutional networks through the transformer-based vision architectures that now compete with and complement them, the ensemble-learning literature underlying structured predictive decision-making, and the neuro-symbolic and explainable-AI literature that has emerged specifically to address the interpretability and reasoning limitations pure connectionist architectures exhibit. The review synthesizes foundational deep learning architecture papers, the ensemble-methods literature spanning bagging, boosting, and stacking, and the more recent neuro-symbolic integration and human-AI decision-boundary literature that shapes how predictive models are deployed for consequential decisions rather than benchmark performance alone. Distinct comparative tables map computer vision architecture families onto their core mechanism and reported benchmark performance, set ensemble and hybrid predictive methods against the specific data structure and decision context each is best suited to address, and cross-reference next-generation CSE application domains against the specific combination of vision, ensemble, and symbolic-reasoning components each domain's deployed systems draw upon. The paper concludes that the field's dominant architectural trajectory is convergent rather than substitutive, with hybrid systems combining perception, structured prediction, and symbolic constraint-checking outperforming any single paradigm applied in isolation, and identifies the systematic characterization of hybrid-system failure modes across component boundaries as the central future research prospect.