Urban　traffic　that　is　heterogeneous　significantly　impacts　on　urban　air　quality　in　both　temporal　and　spatial　scale,　while　traditional　dispersion　models　struggle　to　assess　it　at　high　temporal　resolution　and　multiple　spatial　scales.　This　paper　tried　to　comprehensively　review　the　machine　learning　(ML)-based　high-resolution　traffic-air　quality　(TAQ)　models　and　provide　valuable　insights　for　the　development　and　application.　The　advancements　in　ML-based　TAQ　models　are　highlighted　via　our　analysis　of　103　studies　from　2013　to　2023,　particularly　for　European　regions　where　relevant　research　have　increased　significantly.　The　review　summarized　the　prediction　of　urban　air　quality　influenced　by　complicated　on-road　traffic　conditions　using　various　ML　algorithms　(e.g.,　tree-based　and　neural　network　algorithms).　Additionally,　we　explored　the　sources　of　input　datasets,　feature　applications　and　challenges　associated　with　ML　algorithms’　selection　and　application.　Additionally,　prospects　were　proposed　for　prioritizing　interpretability　in　ML　algorithms　and　the　optimizing　input　metadata　to　improve　the　reliability　and　performance　of　ML-based　TAQ　models.　Together　with　future　research　directions　were　also　discussed,　including　real-time　urban　air　quality　evaluation　and　models　facilitated　concerned　with　health　and　economic　effects.　Aiming　to　advance　ML-based　TAQ　models　in　this　field,　this　review　illustrates　a　brand-new　roadmap　for　understanding　the　intricate　relationship　between　urban　on-road　traffic　and　air　quality　dynamics.　©　2025