Third-Generation Advanced Machine Learning Models for Predicting Financial Asset Returns: Reconciling Predictive Efficiency with Explainability, Algorithmic Fairness, Governance and Cybersecurity. A Systematic Literature Review.

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This systematic review assesses the extent to which third-generation machine learning architectures, both monolithic and hybrid, improve the predictive efficiency of financial asset returns, and examines the conditions under which such performance can be reconciled with the requirements of responsible artificial intelligence. The scope covers efficient Transformers, graph neural networks, deep reinforcement learning, transfer learning, meta-learning, self-supervised learning, explainable and neurosymbolic AI, and causal modelling, across peer-reviewed literature published between January 2018 and March 2026. The protocol follows PRISMA 2020 and PRISMA-P. Four responsible-AI dimensions are examined jointly with performance: explainability, algorithmic fairness, governance and cybersecurity. Expected outputs: a taxonomy of architectures, a comparative performance synthesis by algorithmic family, a risk-of-bias matrix specific to machine learning in finance, a cross-mapping of family against responsibility dimension, and a prioritised research agenda.