A systematic review and unified multi-perspective conceptual synthesis of contemporary machine learning paradigms

Background Machine learning (ML) has advanced many fields, enabling innovations once beyond technological reach. Self-driving cars, real-time language translation, personalized e-commerce, and sophisticated medical diagnostics exemplify its transformative potential. Although ML has become increasingly interdisciplinary, the existing literature often concentrates on individual paradigms or domains. ML cannot be confined to traditional tasks or architectures, given the rise of self-supervised learning, foundation models, federated learning, and growing ethical concerns about data privacy, algorithmic bias, fairness, interpretability, and sustainability. To address this gap, this systematic literature review (SLR) synthesizes contemporary ML developments through a structured, multi-perspective conceptual framework integrating representation, computational nature, learning paradigm, training philosophy, linearity, functional objectives, abstraction layers, hybridization, and special-purpose modeling across application domains. Methodology This review analyzes 81 peer-reviewed studies published between 2017 and 2024 across healthcare, finance, education, engineering, and infrastructure. A PRISMA-guided selection process with explicit inclusion and exclusion criteria and structured keyword-based search strategies was employed to ensure methodological rigor and transparency. Results The review categorizes contemporary ML paradigms across multiple complementary perspectives, highlighting their diverse cross-domain applications. The findings demonstrate a progression from traditional supervised, unsupervised, and reinforcement learning toward advanced paradigms such as self-supervised learning, foundation models, federated learning, and hybrid systems. Cross-domain analysis reveals strong concentration in healthcare and finance, while other societal domains remain underexplored. The study identifies comparative strengths, structural limitations, and persistent challenges related to interpretability, fairness, scalability, generalizability, and responsible deployment. Conclusions This SLR consolidates the evolving ML landscape by synthesizing multiple taxonomic perspectives within a unified framework and demonstrates how representation, computational structure, learning philosophy, abstraction level, and application context shape paradigm selection and deployment. The review highlights structural challenges, including the absence of consistent paradigm definitions, limited cross-paradigm comparisons, and insufficient empirical validation. Fairness, interpretability, privacy, scalability, and sustainability emerge as essential considerations for responsible ML development, particularly in healthcare and finance. Although limited by its timeframe and scope, the proposed taxonomy offers a coherent conceptual foundation and supports systematic, responsible, and theoretically grounded advancement of the field.