<p>Trip generation modelling is the foundational step in travel demand forecasting. Traditional approaches such as multiple linear regression (MLR) and cross-classification face limitations in capturing nonlinear relationships among socioeconomic, built environment, and travel behaviour variables. This review systematically evaluates both traditional and machine learning (ML) techniques for household trip generation modelling. A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) compliant methodology was adopted, screening 3,845 records from four academic databases, resulting in 70 studies included in the final analysis. The reviewed ML techniques include Artificial neural networks (ANNs), ensemble methods, Support vector machines, deep learning architectures, and hybrid fuzzy-neural systems across diverse geographic contexts. The findings indicate that the superiority of ML over traditional methods is context-dependent rather than universal. ANNs outperform MLR primarily in single-city, household-level studies with moderate sample sizes in developing-country contexts. However, ensemble methods have shown near-equivalence with linear regression in geographically heterogeneous, multi-context datasets. Deep learning architectures such as Graph Neural Networks and Convolutional Neural Network–Multidimensional Long Short-Term Memory (CNN-MDLSTM) models achieve high accuracy for spatial trip generation but are limited to scenarios with large-scale datasets and substantial computational resources. Classical MLR remains appropriate where institutional accountability, policy transparency, or resource constraints preclude complex models. The review identifies a complexity–transferability trade-off, where simpler models transfer more reliably across cities. Model selection should therefore be guided by the planning context rather than algorithmic sophistication alone.</p>