Data-Driven Selection of Concrete Modification Strategies: A Critical Review of Fibres, Polymers, Waste Materials and Machine-Learning Tools for Mechanical, Durability and Structural Performance

Concrete can now be modified in many ways, but engineers still lack a clear method for choosing between the options.Fibres, polymer admixtures, waste-derived materials and supplementary cementitious materials each act through a different mechanism, and each improves some properties while weakening others.This review treats modification as a selection problem rather than a list of additives.For each strategy it identifies the property it controls, the mechanism behind the effect, the main trade-off and the minimum evidence needed to support a performance claim.Mechanical, durability and structural performance are considered as separate selection criteria, because a mix that performs well in one may fail in another.The role of machine learning is then examined as a tool for screening and ranking candidate mixtures before testing.The literature shows that fibres mainly improve post-cracking behaviour and impact resistance, polymers improve adhesion and reduce permeability at some cost in compressive strength, and waste and supplementary materials reduce clinker demand but require longer curing and careful characterisation.Machine-learning models predict compressive strength well within their training range, but datasets for modified mixtures, durability and structural response remain small and poorly described.A five-step data-driven selection framework and a property-strategy matrix are proposed.The review concludes that machine learning is most useful for narrowing the choice of mixtures, while final selection must still rest on targeted experiments.