AI for Intelligent Transportation Systems: A Systematic Review of Applications in Demand-Responsive Transport.

Demand-Responsive Transport (DRT) has emerged as a flexible public transport strategy to improve accessibility, service coverage, and operational efficiency in contexts where conventional fixed-route services are inefficient, insufficient, or difficult to operate. In parallel, Artificial Intelligence (AI), and particularly Machine Learning (ML), has increasingly been investigated for predictive tasks relevant to DRT planning and operations, including demand forecasting, travel-time estimation, service reliability assessment, and decision support. This study presents a systematic literature review of ML-based predictive analytics in public-transport-oriented DRT services. After the screening and eligibility process, the included studies were analyzed according to predictive task, data source, modelling technique, validation strategy, performance metrics, and implementation-related challenges; methodological quality, risk of bias (RoB), and applicability were assessed using a PROBAST+AI-based framework. The results show that research in this field has expanded rapidly in recent years and is moving from isolated demand prediction models towards more integrated frameworks linking prediction, optimization, and service planning. However, the evidence remains fragmented, with substantial heterogeneity in data sources, spatial and temporal scales, modelling approaches, and evaluation procedures. The quality assessment showed generally favorable predictor quality and low outcome-related RoB, but analysis-related RoB was high in just over half of the studies, mainly because independent evaluation and validation accounting for temporal, spatial, or simulation-induced dependence were often lacking. Most studies provided retrospective, offline, simulation-based, or conceptual decision-support evidence, whereas prospective field deployment, external validation, and post-implementation monitoring were rarely or insufficiently documented. This review therefore provides a structured synthesis of current ML research in DRT and identifies priorities for future work, including improved reproducibility, stronger validation, robust baseline comparisons, multiple evaluation metrics, greater interpretability, and prospective assessment under real-world operational conditions.